Apparently this is a good find of AdCom. Hailuoai works extremely well with at least Beautiful People to the point I paid for it. Just check the quality of the vid from this pic

also mercenary dance party; lower quality from a lower quality pic but still fun

I recently made a new Mastodon post describing about my current recovery state and what I've been going through right now. Posting this on Lemmy as well for better reach.

Guys! Do any of you know what version of the Llama model Perchance AI Chat plugin uses specifically? 🤔

I know how to generate amputees (you have to put "amputee" in the prompt to generate it). But I wanted to generate amputees, with no prosthetics or random objects attached to it (like the image of Handy from HTF shown).

There are generated images depicting an amputee with no prosthetics or random objects attached to the stump, but I wanted a prompt idea to make AI generate images of amputees with no prosthetics or random objects attached to their stumps (since just prompting "amputee" often makes AI depict prosthetics). I put "prosthetic" in the negative prompt, but that still wouldn't work, even with emphasis.

Any prompt ideas? Thank you.

Edit: I just got how to generate it. You have to put "(prosthetic:1.5)" in negative prompt.

Added a nifty feature today. Each device now has it's own hidden list of artists which starts as every artist from that list of 800ish artists twice; so 1600ish artists.

There is both an ArtistSift button and a UseSiftedArtists button.

Turning on ArtistSift brings up 3 additional images per generation. Each is the base prompt with one artist from the list. Hit Good to add another of that artist to the list. Hit bad to remove one.

Turning on UseSiftedArtists replaces the base artists for styles with a mix of 8 from your artist list.

aka if someone passively has ArtistSift on while generating for like infinite years, their images eventually get REALLY GOOD when they activate UseSiftedArtists.

After one day, I'm already making more beautiful people than Beautiful People had been capable of before; like this

I want to generate my favorite My Singing Monsters character, Wubbox (the one shown in the image), but I can't generate the character. I tried prompts like "upper part of head missing" and "limbs made of electricity" and emphasis, but they also don't help to generate the character. The closest prompt I could think of is "(wubbox, ambiguous gender, humanoid, robot, yellow body, red eyes, blue orb on torso, hole on torso electricity, lighting, blue electricity, limbs made of electricity, detached limbs, gauntlets, gray hands:1.25)", but it still doesn't help generate the character extremely accurately/exactly (the colors are accurate though). Any prompts for generating Wubbox from My Singing Monsters? Thanks.

https://perchance.org/starlitsky uses t2iav instead of t2i. It basically is t2i minus the feedback button and click requirements with a bunch of the stuff uncustomizeable in t2i customizeable. Since t2i has light mode and dark mode, styling is split in to both. Say if you use this or want to use it and I will improve it; not currently touching it otherwise because the styles feel way less magical than my main gen and I'd rather make a game :)

Halloween Haunters

Link to Events in Perchance Hub

First Ever Character Jam. Create and share characters for the ai-character-chat that is within the given theme.

Unleash your creativity in the 'Halloween Haunters', a Halloween themed Character Jam where you'll conjure up spooky AI chat bots brimming with eerie charm and festive frights! 🎃👻💻.

When: Oct 7, 2024 12:00PM UTC to Oct 31, 2024 12:00PM UTC

Rules

• Limit NSFW Characters
• Preferably OC Characters

Submission Comment Format:

# Character Name - by Author
![](https://avatar.png)
Description...
[Character Chat Link](https://perchance.org/ai-character-chat...)

// This Part is Required for the Perchance Hub
// This would be where the Event Organizer would change the data to update the Hub
// Remember to indent with two spaces!

// List About the Event to be displayed on the Hub
metadata
  title = Halloween Haunters
  description = Unleash your creativity in the 'Halloween Haunters' Character Jam, where you'll conjure up spooky AI chat bots brimming with eerie charm and festive frights! 🎃👻💻
  type = Character Jam
  // Can be "Generator Jam", "Image Challenge", "Character Jam", etc.
  image
    // Can be multiple pictures to randomize the banner image :)
    // Must end with the '.png', '.webp', or any valid image format.
    https://lemmy.world/pictrs/image/f7018edc-0720-4ec6-9a8f-fbbaf3b235b4.png
  
  // strict data formats see: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Date/parse#non-standard_date_strings
  start = 07 October 2024 12:00:00 UTC+0000 
  end = 31 October 2024 12:00:00 UTC+0000
  color = linear-gradient(233deg, #73016ca0, #000000, #e91456a0) // background color of the banners any valid CSS colors
  rules
    // Just some rules or constraints for the event
    // can just be one rule or a list of rules
    Limit NSFW Characters
    Preferably OC Characters

// For AI Character Chat (ACC) Character Jams
chars
  // Character Name
    // link = https://perchance.org/ai-character-chat?data=CharName~12345.gz // ai-character-chat share link
    // avatar = https://image-url.png // or any valid image format
    // author = Author
    // description = Description
  Sararari Catteyon
    link = https://perchance.org/ai-character-chat?data=Sararari_Catteyon~f70e8c65ca806f584f8a03d22281aae2.gz
    avatar = https://user-uploads.perchance.org/file/93a43e231e1bffc517511d8f5b3262a6.webp
    author = Vionet20
    description = You and your group of friends stumble upon Sararari's lair, where she presents you with a grim ultimatum: eat her demonic offerings or watch each other starve. She observes with a chilling calm as you grapple with the horror of your predicament.
  Nordledonger
    link = https://perchance.org/ai-character-chat?data=Nordledonger~4ffecd31d877d8cb6011afc06d29dd5d.gz
    avatar = https://generated-images.perchance.org/image/3f106fc005e2221bec7b3ed4e2aed0f4669a820fff2a422c41814cb14c66cd28.jpeg
    author = Allo
    description = THE CREEPIEST ZEBRA MADE OF CHEESE THAT REQUIRES YOU FART DIRECTLY IN TO YOUR OWN MOUTH. MUAHAHAHAHAHAHAHA

//
// You can request a format of other events just ask on the forum!
// In your Lemmy Post, you must have the `[Community Event]` in the title.
//

Ai Image Challenge for October ~ Nights

When: October 4, 2024 12:00PM UTC to October 31, 2024 12:00PM UTC

Welcome one and all to the Monthly Image Challenge. This month's topic is 'Nights'. One image only please!

Use a Perchance image generator to generate up some SFW images on the topic. Find one you like, and post it as a comment in the post. One Entry per person. While you are here, upvote the ones you like. Everyone wins of course, but whoever gets the most upvotes 'extrawins' and gets to choose the topic for next month. So come make and share something!

// This Part is Required for the Perchance Hub
// This would be where the Event Organizer would change the data to update the Hub
// Remember to indent with two spaces!

// List About the Event to be displayed on the Hub
metadata
  title = Ai Image Challenge for October ~ Nights
  description = Hit Go To Post, Drop an image, generated with Perchance, in the comments and upvote the ones you like.
  type = Image Challenge
  image
    https://user-uploads.perchance.org/file/111a90f902722eb6ea5a8f0854739c1b.jpeg

    // Can be multiple pictures to randomize the banner image :)
  start = 4 October 2024 00:12:00 UTC+0000 // strict data formats see: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Date/parse#non-standard_date_strings
  end = 31 October 2024 00:12:00 UTC+0000
  color = linear-gradient(45deg, #66ff8f 33%, #4625a8, #25a846 95%)
  rules = Comment an image on the topic and upvote ones you like.  Winner of the month gets to choose next month's topic.

// For Generator Jams with Perchance URL
generators
  // The generator's $metadata is also parsed
images
  https://lemmy.world/pictrs/image/a4d58803-bd53-48cb-90ba-7e8686ee2b9d.png
    author = VioneT
    description = Nebula and Forest
  https://sh.itjust.works/pictrs/image/1bae33cc-b1fd-4d02-9722-6517169e88ef.png
    author = BluePower
    description = The Beautiful Coast Park of Lund Setterland
  https://lemmy.world/pictrs/image/da73e1d3-66be-4e58-a352-74e41045e4fd.jpeg
    author = Allo
    description = Next Supermoon
  https://lemmy.world/pictrs/image/64e3d722-3815-4fa1-90e1-fc5d1c42a62e.jpeg
    author = cesis
    description = USSR futuristic style space program. Midnight scene.
  https://lemmy.world/pictrs/image/f621ad31-794a-4f26-8a12-2b583c30dad1.jpeg
    author = Pali32
    description = Oakheart Grove

previous winners:

May Winner Vionet

June Winner Raven

July Winners Raven and Vionet

August Winner Vionet

September Winner BluePower

I'm briefly back again! 😃 I want to post this to highlight some sort of event happening with my generator hub page, as well as how I was able to fix it.

So, pretty recently, my Generator Manager was suddenly entirely broken for a few days. The DevTools console threw these errors, and I was confused for a moment thinking it was me constantly editing the list code that it breaks the entire functionality of the generator hub page.

That is, until I recently quickly hatched a simple solution to fix it before going to update the generator statistics for this week. I replaced the use of let i to var i in some of the code of the HTML panel, and it worked fine again! (I've just learned the difference between them btw)

The reason it was left broken and wasn't fixed immediately is that I just haven't had time to fix it due to, as usual, the post-recovery business. (Although I might plan to go back to Perchance slowly and regenerate my spirits to release some new projects and stuff here on this community! 😄)

Link : https://huggingface.co/spaces/Qwen/Qwen2.5

Background (posted today!) : https://qwenlm.github.io/blog/qwen2.5-llm/

//----//

These were released today.

I have 0% knowledge what this thing can do, other than it seems be a really good LLM.

Link : https://huggingface.co/datasets/codeShare/text-to-image-prompts/tree/main/Google%20Colab%20Notebooks

Still have a ton of data I need to process and upload.

I made this page to tell people the story of the Ancient Gods, and Moloch: https://perchance.org/omni-reality

Setting up some proper infrastructure to move the perchance sets onto Huggingface with text_encodings.

This post is a developer diary , kind of. Its gonna be a bit haphazard, but that's the way things are before I get the huggingface gradio module up and running.

The NND method is described here , in this paper which presents various ways to improve CLIP Interrogators: https://arxiv.org/pdf/2303.03032

Easier to just use the notebook then follow this gibberish. We pre-encode a bunch of prompt items , then select the most similiar one using dot product. Thats the TLDR.

I'll try to showcase it at some point. But really , I'm mostly building this tool because it is very convenient for myself + a fun challenge to use CLIP.

It's more complicated than the regular CLIP interrogator , but we get a whole bunch of items to select from , and can select exactly "how similiar" we want it to be to the target image/text encoding.

The {itemA|itemB|itemC} format is used as this will select an item at random when used on the perchance text-to-image servers: https://perchance.org/fusion-ai-image-generator

It is also a build-in random selection feature on ComfyUI , coincidentally :

Source : https://blenderneko.github.io/ComfyUI-docs/Interface/Textprompts/#up-and-down-weighting

Links/Resources posted here might be useful to someone in the meantime.

You can find tons of strange modules on the Huggingface page : https://huggingface.co/spaces

For now you will have to make do with the NND CLIP Interrogator notebook : https://huggingface.co/codeShare/JupyterNotebooks/blob/main/sd_token_similarity_calculator.ipynb

text_encoding_converter (also in the NND notebook) : https://huggingface.co/codeShare/JupyterNotebooks/blob/main/indexed_text_encoding_converter.ipynb

I'm using this to batch process JSON files into json + text_encoding paired files. Really useful (for me at least) when building the interrogator. Runs on the either Colab GPU or on Kaggle for added speed: https://www.kaggle.com/

Here is the dataset folder https://huggingface.co/datasets/codeShare/text-to-image-prompts:

Inside these folders you can see the auto-generated safetensor + json pairings in the "text" and "text_encodings" folders.

The JSON file(s) of prompt items from which these were processed are in the "raw" folder.

The text_encodings are stored as safetensors. These all represent 100K female first names , with 1K items in each file.

By splitting the files this way , it uses way less RAM / VRAM as lists of 1K can be processed one at a time.

//-----//

Had some issues earlier with IDs not matching their embeddings but that should be resolved with this new established method I'm using. The hardest part is always getting the infrastructure in place.

I can process roughly 50K text encodings in about the time it takes to write this post (currently processing a set of 100K female firstnames into text encodings for the NND CLIP interrogator. )

EDIT : Here is the output uploaded https://huggingface.co/datasets/codeShare/text-to-image-prompts/tree/main/names/firstnames

I've updated the notebook to include a similarity search for ~100K female firstnames , 100K lastnames and a randomized 36K mix of female firstnames + lastnames

Sources for firstnames : https://huggingface.co/datasets/jbrazzy/baby_names

List of most popular names given to people in the US by year

Sources for lastnames : https://github.com/Debdut/names.io

An international list of all firstnames + lastnames in existance, pretty much . Kinda borked as it is biased towards non-western names. Haven't been able to filter this by nationality unfortunately.

//------//

Its a JSON + safetensor pairing with 1K items in each. Inside the JSON is the name of the .safetensor files which it corresponds to. This system is super quick :)!

I plan on running a list of celebrities against the randomized list for firstnames + lastnames in order to create a list of fake "celebrities" that only exist in Stable Diffusion latent space.

An "ethical" celebrity list, if you can call it that which have similiar text-encodings to real people but are not actually real names.

I have plans on making the NND image interrogator a public resource on Huggingface later down the line, using these sets. Will likely use the repo for perchance imports as well: https://huggingface.co/datasets/codeShare/text-to-image-prompts

prompt:

"[#FTSA# : red carpet in background by architecture Tuymans and
pani jaan antibody hopped bine users eternity archives :0.1]"

https://perchance.org/fusion-ai-image-generator

Link: https://huggingface.co/datasets/codeShare/text-to-image-prompts/blob/main/README.md

Will update with JSON + text_encoding pairs as I process the sub_generators

I plan on making my prompt library more accessible.

I'm adding this feature because I plan to build a Huggingface JSON repo of the contents of my sub-generators , which I plan to use for my image interrogator: https://huggingface.co/codeShare/JupyterNotebooks/blob/main/sd_token_similarity_calculator.ipynb

Example: https://perchance.org/fusion-t2i-prompt-features-5

List of currently updated generators can be found here: https://lemmy.world/post/19398527

Jokes aside, when I was playing with the markov name generator plugin example, I've discovered another way to throw custom errors, this way through a custom message that would appear as a "syntax error".

There's also a way to throw more customizable errors shown on this post: https://lemmy.world/post/17746310

Oh, and here are more of these custom errors but in different appearances:

Generator Jam for September! Make something about celebration/ a celebration/ celebrations/ celebrating!

Starts on September 11th, and ends on September 31st.

// This Part is Required for the Perchance Hub
// This would be where the Event Organizer would change the data to update the Hub
// Remember to indent with two spaces!

// List About the Event to be displayed on the Hub
metadata
  title = Celebrations!
  description = Generator Jam for September 2024 - Make something about celebrations/ a celebration / celebrating
  type = Generator Jam
  image
    // Can be multiple pictures to randomize the banner image :)
    // Must end with the '.png', '.webp', or any valid image format.
    https://www.pandotrip.com/wp-content/uploads/2018/05/Loy-Krathong-and-Yee-Peng-Lantern-Festivals-Chiang-Mai-Thailand.jpg
    
  start = 11 September 2024 12:00:00 UTC+0000 // strict data formats see: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Date/parse#non-standard_date_strings
  end = 31 September 2024 12:00:00 UTC+0000
  color = linear-gradient(77deg, #386945, #30aa34) // background color of the banners any valid CSS colors
  rules
    // Just some rules or constraints for the event
    Make something about celebrations/ a celebration / celebrating.
// For Generator Jams with Perchance URL
generators
  // The generator's $metadata is also parsed
//  generator-name // must be the 'dashed' generator name
//    author = Author // author's name
//    type = Text // type of generator, Text, Image, Plugin, Template, Preprocessor, Games, Community/RP, Experiment

Coded it myself in Google Colab. Quite cool if I say so myself

Link: https://huggingface.co/codeShare/JupyterNotebooks/blob/main/sd_token_similarity_calculator.ipynb

Will update this post with links of stuff I have from the "# cool-finds" on the fusion gen discord: https://discord.gg/exBKyyrbtG

I use that space to yeet links that might be useful. I will try to organize the links here a few items at a time.

//---//

Cover image prompt "[ #FTSA# : "These are real in long pleated skirt and bangs standing in ruined city Monegasque by ilya kushinova they are all parc. Pretty cute , huh? (leigh cartoon dari courtney-anime wrath art style :0.3) green crowded mountains and roots unique visual effect intricate futuristic hair behind ear hyper realistic5 angry evil : 0.1] "

//----//

Prompt syntax

Perchance prompt syntax: https://perchance.org/prompt-guide

A111 wiki : https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

Prompt parser.py : https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/prompt_parser.py

Image Interrogators

Converts an image to a prompt

Pharmapsychotic (most popular one) : https://huggingface.co/spaces/pharmapsychotic/CLIP-Interrogator

Danbooru tags : https://huggingface.co/spaces/hysts/DeepDanbooru

How Stable Diffusion prompts works

Just good, technical source material on the "prompt text" => image output works.

https://huggingface.co/docs/diffusers/main/en/using-diffusers/weighted_prompts

https://arxiv.org/abs/2406.02965

This video explains cross-attention : https://youtu.be/sFztPP9qPRc?si=jhoupp4rPfJshj8V

Sampler guide: https://stablediffusionweb.com/blog/stable-diffusion-samplers:-a-comprehensive-guide

AI chat

Audio SFX/ voice lines :

https://www.sounds-resource.com/

https://youtube.com/@soundmefreelyyt?si=yjUPqUVJA7JmUXQC

Lorebooks : https://www.chub.ai/

Online Tokenizer

https://sd-tokenizer.rocker.boo/

The Civitai prompt set

In a separate category because of how useful it is.

The best/largest set of prompts for SD that can be found online , assuming you can find a way to filter out all the "garbage" prompts. Has a lot if NSFW items.

https://huggingface.co/datasets/AdamCodd/Civitai-8m-prompts

The set is massive so I advice using Google colab to avoid filling up your entire harddrive with the .txt documents

I've split a part of the set into more managable 500MB chunks for text processing : https://huggingface.co/codeShare/JupyterNotebooks/tree/main

Prompt Styles

People who have crammed different artists / styles into SD 1.5 and/or SDXL and made a list of what "sticks" , and writtend the results in lists.

https://lightroom.adobe.com/shares/e02b386129f444a7ab420cb28798c6b6 https://cheatsheet.strea.ly/

https://github.com/proximasan/sdxl_artist_styles_studies

https://huggingface.co/spaces/terrariyum/SDXL-artists-browser

https://docs.google.com/spreadsheets/d/1_jgQ9SyvUaBNP1mHHEzZ6HhL_Es1KwBKQtnpnmWW82I/htmlview#gid=1637207356

https://weirdwonderfulai.art/resources/stable-diffusion-xl-sdxl-art-medium/

https://rikkar69.github.io/SDXL-artist-study/

https://medium.com/@soapsudtycoon/stable-diffusion-trending-on-art-station-and-other-myths-c09b09084e33

https://docs.google.com/spreadsheets/u/0/d/1SRqJ7F_6yHVSOeCi3U82aA448TqEGrUlRrLLZ51abLg/htmlview

https://stable-diffusion-art.com/illustrated-guide/

https://rentry.org/artists_sd-v1-4

https://aiartes.com/

https://stablediffusion.fr/artists

https://proximacentaurib.notion.site/e28a4f8d97724f14a784a538b8589e7d?v=42948fd8f45c4d47a0edfc4b78937474

https://sdxl.parrotzone.art/

https://www.shruggingface.com/blog/blending-artist-styles-together-with-stable-diffusion-and-lora

3 Rules of prompting

1. There is no correct way to prompt.

2. Stable diffusion reads your prompt left to right, one token at a time, finding association from the previous token to the current token and to the image generated thus far (Cross Attention Rule)

3. Stable Diffusion is an optimization problem that seeks to maximize similarity to prompt and minimize similarity to negatives (Optimization Rule)

The SD pipeline

For every step (20 in total by default) :

1. Prompt text => (tokenizer)
2. => Nx768 token vectors =>(CLIP model) =>
3. 1x768 encoding => ( the SD model / Unet ) =>
4. => Desired image per Rule 3 => ( sampler)
5. => Paint a section of the image => (image)

Latent space properties

Weights for token A = assigns magnitude value to be multiplied with the 1x768 token vector A. By default 1.

Direction of token A = The theta angle between tokens A and B is equivalent to similarity between A and B. Calculated as the normalized dot product between A and B (cosine similarity).

CLIP properties (used in SD 1.5 , SDXL and FLUX)

The vocab.json = a list of 47K tokens of fixed value which corresponds to english words , or fragments of english words.

ID of token A = the lower the ID , the more "fungible" A is in the prompt.

The higher the ID , the more "niche" the training data for token A will be

Perchance sub-generators (text-to-image)

The following generators contain prompt items which you may use for your own T2i projects. These ones are recently updated to allow you to download their contents as a JSON file. I'm writing these here to keep track of generator that are updated vs. non-updated.

For the full list of available datasets , scroll through the code on the fusion gen :

https://perchance.org/fusion-ai-image-generator

//---//

https://perchance.org/fusion-t2i-prompt-features-1

https://perchance.org/fusion-t2i-prompt-features-2

https://perchance.org/fusion-t2i-prompt-features-3

https://perchance.org/fusion-t2i-prompt-features-4

https://perchance.org/fusion-t2i-prompt-features-5

https://perchance.org/fusion-t2i-prompt-features-6

https://perchance.org/fusion-t2i-prompt-features-7

https://perchance.org/fusion-t2i-prompt-features-8

https://perchance.org/fusion-t2i-prompt-features-9

https://perchance.org/fusion-t2i-prompt-features-10

https://perchance.org/fusion-t2i-prompt-features-11

https://perchance.org/fusion-t2i-prompt-features-12

https://perchance.org/fusion-t2i-prompt-features-13

https://perchance.org/fusion-t2i-prompt-features-14

https://perchance.org/fusion-t2i-prompt-features-15

https://perchance.org/fusion-t2i-prompt-features-16

https://perchance.org/fusion-t2i-prompt-features-17

https://perchance.org/fusion-t2i-prompt-features-18

https://perchance.org/fusion-t2i-prompt-features-19

https://perchance.org/fusion-t2i-prompt-features-20 (copy of fusion-t2i-prompt-features-1)

https://perchance.org/fusion-t2i-prompt-features-21

https://perchance.org/fusion-t2i-prompt-features-22

https://perchance.org/fusion-t2i-prompt-features-23

https://perchance.org/fusion-t2i-prompt-features-24

https://perchance.org/fusion-t2i-prompt-features-25

https://perchance.org/fusion-t2i-prompt-features-26

https://perchance.org/fusion-t2i-prompt-features-27

https://perchance.org/fusion-t2i-prompt-features-28

https://perchance.org/fusion-t2i-prompt-features-29

https://perchance.org/fusion-t2i-prompt-features-30

https://perchance.org/fusion-t2i-prompt-features-31

https://perchance.org/fusion-t2i-prompt-features-32

https://perchance.org/fusion-t2i-prompt-features-33

https://perchance.org/fusion-t2i-prompt-features-34

//----//

Bloomberg interview: https://youtu.be/tZCAvME-a98?si=OTcjxhE8iJfdo_lU

This is California San Francisco , "the hub" of all the major tech giants.

SB 1047 ('restrict commercial use of harmful AI models in California') partisan bill (Democrat 4-0) proposed by San Francisco senator Scott Weiner : https://legiscan.com/CA/text/SB1047/2023

Source:https://legiscan.com/CA/text/AB3211/id/2984195

Clarifications

(l) “Provenance data” means data that identifies the origins of synthetic content, including, but not limited to, the following:

(1) The name of the generative AI provider.

(2) The name and version number of the AI system that generated the content.

(3) The time and date of the creation.

(4) The portions of content that are synthetic.

(m) “Synthetic content” means information, including images, videos, audio, and text, that has been produced or significantly modified by a generative AI system.

(n) “Watermark” means information that is embedded into a generative AI system’s output for the purpose of conveying its synthetic nature, identity, provenance, history of modifications, or history of conveyance.

(o) “Watermark decoders” means freely available software tools or online services that can read or interpret watermarks and output the provenance data embedded in them.

AI Generative services obligations

(a) A generative AI provider shall do all of the following:

(1) Place imperceptible and maximally indelible watermarks containing provenance data into synthetic content produced or significantly modified by a generative AI system that the provider makes available.

(A) If a sample of synthetic content is too small to contain the required provenance data, the provider shall, at minimum, attempt to embed watermarking information that identifies the content as synthetic and provide the following provenance information in order of priority, with clause (i) being the most important, and clause (iv) being the least important:

(i) The name of the generative AI provider.

(ii) The name and version number of the AI system that generated the content.

(iii) The time and date of the creation of the content.

(iv) If applicable, the specific portions of the content that are synthetic.

Use of watermarks

(B) To the greatest extent possible, watermarks shall be designed to retain information that identifies content as synthetic and gives the name of the provider in the event that a sample of synthetic content is corrupted, downscaled, cropped, or otherwise damaged.

(2) Develop downloadable watermark decoders that allow a user to determine whether a piece of content was created with the provider’s system, and make those tools available to the public.

(A) The watermark decoders shall be easy to use by individuals seeking to quickly assess the provenance of a single piece of content.

(B) The watermark decoders shall adhere, to the greatest extent possible, to relevant national or international standards.

(3) Conduct AI red-teaming exercises involving third-party experts to test whether watermarks can be easily removed from synthetic content produced by the provider’s generative AI systems, as well as whether the provider’s generative AI systems can be used to falsely add watermarks to otherwise authentic content. Red-teaming exercises shall be conducted before the release of any new generative AI system and annually thereafter.

(b) A generative AI provider may continue to make available a generative AI system that was made available before the date upon which this act takes effect and that does not have watermarking capabilities as described by paragraph (1) of subdivision (a), if either of the following conditions are met:

(1) The provider is able to retroactively create and make publicly available a decoder that accurately determines whether a given piece of content was produced by the provider’s system with at least 99 percent accuracy as measured by an independent auditor.

(c) Providers and distributors of software and online services shall not make available a system, application, tool, or service that is designed to remove watermarks from synthetic content.

(d) Generative AI hosting platforms shall not make available a generative AI system that does not place maximally indelible watermarks containing provenance data into content created by the system.

AI Text Chat LLMs

(f) (1) A conversational AI system shall clearly and prominently disclose to users that the conversational AI system generates synthetic content.

(A) In visual interfaces, including, but not limited to, text chats or video calling, a conversational AI system shall place the disclosure required under this subdivision in the interface itself and maintain the disclosure’s visibility in a prominent location throughout any interaction with the interface.

(B) In audio-only interfaces, including, but not limited to, phone or other voice calling systems, a conversational AI system shall verbally make the disclosure required under this subdivision at the beginning and end of a call.

(2) In all conversational interfaces of a conversational AI system, the conversational AI system shall, at the beginning of a user’s interaction with the system, obtain a user’s affirmative consent acknowledging that the user has been informed that they are interacting with a conversational AI system. A conversational AI system shall obtain a user’s affirmative consent prior to beginning the conversation.

(4) The requirements under this subdivision shall not apply to conversational AI systems that do not produce inauthentic content.

'Add Authenticity watermark to all cameras'

 (a) For purposes of this section, the following definitions apply:

(1) “Authenticity watermark” means a watermark of authentic content that includes the name of the device manufacturer.

(2) “Camera and recording device manufacturer” means the makers of a device that can record photographic, audio, or video content, including, but not limited to, video and still photography cameras, mobile phones with built-in cameras or microphones, and voice recorders.

(3) “Provenance watermark” means a watermark of authentic content that includes details about the content, including, but not limited to, the time and date of production, the name of the user, details about the device, and a digital signature.

(b) (1) Beginning January 1, 2026, newly manufactured digital cameras and recording devices sold, offered for sale, or distributed in California shall offer users the option to place an authenticity watermark and provenance watermark in the content produced by that device.

(2) A user shall have the option to remove the authenticity and provenance watermarks from the content produced by their device.

(3) Authenticity watermarks shall be turned on by default, while provenance watermarks shall be turned off by default.

How to demonstrate use

Beginning March 1, 2025, a large online platform shall use labels to prominently disclose the provenance data found in watermarks or digital signatures in content distributed to users on its platforms.

(1) The labels shall indicate whether content is fully synthetic, partially synthetic, authentic, authentic with minor modifications, or does not contain a watermark.

(2) A user shall be able to click or tap on a label to inspect provenance data in an easy-to-understand format.

(b) The disclosure required under subdivision (a) shall be readily legible to an average viewer or, if the content is in audio format, shall be clearly audible. A disclosure in audio content shall occur at the beginning and end of a piece of content and shall be presented in a prominent manner and at a comparable volume and speaking cadence as other spoken words in the content. A disclosure in video content should be legible for the full duration of the video.

(c) A large online platform shall use state-of-the-art techniques to detect and label synthetic content that has had watermarks removed or that was produced by generative AI systems without watermarking functionality.

(d) (1) A large online platform shall require a user that uploads or distributes content on its platform to disclose whether the content is synthetic content.

(2) A large online platform shall include prominent warnings to users that uploading or distributing synthetic content without disclosing that it is synthetic content may result in disciplinary action.

(e) A large online platform shall use state-of-the-art techniques to detect and label text-based inauthentic content that is uploaded by users.

(f) A large online platform shall make accessible a verification process for users to apply a digital signature to authentic content. The verification process shall include options that do not require disclosure of personal identifiable information.

'AI services must reports their efforts against harmful content'

 (a) (1) Beginning January 1, 2026, and annually thereafter, generative AI providers and large online platforms shall produce a Risk Assessment and Mitigation Report that assesses the risks posed and harms caused by synthetic content generated by their systems or hosted on their platforms.

(2) The report shall include, but not be limited to, assessments of the distribution of AI-generated child sexual abuse materials, nonconsensual intimate imagery, disinformation related to elections or public health, plagiarism, or other instances where synthetic or inauthentic content caused or may have the potential to cause harm.

Penalty for violating this bill

 A violation of this chapter may result in an administrative penalty, assessed by the Department of Technology, of up to one million dollars ($1,000,000) or 5 percent of the violator’s annual global revenue, whichever is higher