Thats true! Since nat lang text encoders are more complex , the negatives of the stuff being encoded is rarely the opposite.
As in , the model itself (Chroma on perchance) isnt trained to comprehend negative vectors. Tho lodestones (creator of Chroma) never specified but I assume not.
Negatives are an off shoot in training that sorta worked in CLIP based models SD1.5 SDXL (Pony / illustrious) that carried over into the nat lang mod releases out of habit in the community.
It worked in CLIP models because the CLIP encoder is simple. Write 'ice cream' in CLIP and the text encoding vector will point roughly same direction no matter where in the 75 token batch 'ice cream' .
Compare that to the many number of different answers you can get from chatGPT or GROK containing the word 'ice cream' and you can see how the 512 token batch encoding of the T5 in Chroma , or Qwen encoder in Klein / Z image varies drastically depending how common words are arranged in the text.
Perchance chat is gonna get the site shut down I swear.
Any forum with a buncha unhinged weirdos + underage users is trouble waiting to happen for the site. People wanna chat , do that on Lemmy or whatever other moderated forums there are IMO.
2025 Toronto International Film Festival - Black Excellence Brunch TORONTO, ONTARIO - SEPTEMBER 08: (L-R) Karen Chapman, Regina Taylor and Joan Jenkinson attend the Black Excellence Brunch during the 2025 Toronto International Film Festival at Petros82 Restaurant on September 08, 2025 in Toronto, Ontario. (Photo by Leon Bennett/Getty Images)
Thanks. I appreciate the effort and time you put into this. ...But is your keyboard broken or something haha? Where are the punctuations . , ? lol
Fair enough , but some exposition would save you coming off as a total weirdo that posts cryptic messages on off site platform
You are a weird person OP and I have no clue what you are talking about
Thats true! Since nat lang text encoders are more complex , the negatives of the stuff being encoded is rarely the opposite.
As in , the model itself (Chroma on perchance) isnt trained to comprehend negative vectors. Tho lodestones (creator of Chroma) never specified but I assume not.
Negatives are an off shoot in training that sorta worked in CLIP based models SD1.5 SDXL (Pony / illustrious) that carried over into the nat lang mod releases out of habit in the community.
It worked in CLIP models because the CLIP encoder is simple. Write 'ice cream' in CLIP and the text encoding vector will point roughly same direction no matter where in the 75 token batch 'ice cream' .
Compare that to the many number of different answers you can get from chatGPT or GROK containing the word 'ice cream' and you can see how the 512 token batch encoding of the T5 in Chroma , or Qwen encoder in Klein / Z image varies drastically depending how common words are arranged in the text.
Ask the llm
Its gpt-4 by openAI , you can ask the llm model itself who it is https://perchance.org/ai-text-generator
Perchance chat is gonna get the site shut down I swear.
Any forum with a buncha unhinged weirdos + underage users is trouble waiting to happen for the site. People wanna chat , do that on Lemmy or whatever other moderated forums there are IMO.
Tell you what ; If ganking up on strangers online like this is what happens in perchance chats I'm happy the 16 year olds stay clear of it.
Have your internet points. Just keep whatever strange community you have in this chat away from me.
Is nutty behavior seeing a 16 year old on a chat forum and reacting that he should be banned.
Like... what you do is completely crazy behavior.
I'm stopping you in your tracks here bud. Time-out. Is what I wanna say.
Who told you this? Like...why?
You are disagreeing because of your own ego. Stop it.
I'm not looking to claim any points.
You ask for info , I give info. Why so distrustful?
Perchance model is FLUX Chroma.
FLUX Chroma Flash Heun
Easiest way to get photoreal output is to head off to getty and copy a photo caption: https://www.gettyimages.com/editorial-images
Example output on perchance generator
2025 Toronto International Film Festival - Black Excellence Brunch TORONTO, ONTARIO - SEPTEMBER 08: (L-R) Karen Chapman, Regina Taylor and Joan Jenkinson attend the Black Excellence Brunch during the 2025 Toronto International Film Festival at Petros82 Restaurant on September 08, 2025 in Toronto, Ontario. (Photo by Leon Bennett/Getty Images)From : https://www.gettyimages.com/editorial-images/entertainment/event/toronto-international-film-festival-black-excellence-brunch/776375508?editorialproducts=all
I don't mind hearing what you are actually angry about if its something personal.
You are not yourself man.
Take a break and get a grip on reality.
Revert to your mental health checklist of things to do.
Onsite limit is 768x768 but its Perchance T2i model is FLUX Chroma and you can run it for free on Tensor Art and other places at higher resolutions
Are you ok? I hope you are doing well , man
FLUX Chroma (Perchance T2i model) is literally the best model on the market haha.
I've collected what prompt I can find on Chroma training data here (see screenshot above for output): https://huggingface.co/datasets/codeShare/chroma_prompts/blob/main/parquet_explorer.ipynb
So try mimicing that prompt format 👍
Good scientific approach!
I collected training prompts from lodestones repo to get an idea of the prompt format for Chroma: https://huggingface.co/datasets/codeShare/chroma_prompts/blob/main/parquet_explorer.ipynb
Still early stages so you'll have to download the .parquet file to your own Google Drive and access it via the notebook from there