Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/zer0int/comfyui-nuke-a-text-encoder
For SDXL, SD1.5, Flux. Nuke T5 and let CLIP guide Flux.1 on its own! Or let let random guide Flux.1! Or load a CLIP crazy opinion embedding about your image and let that guide the AI!
https://github.com/zer0int/comfyui-nuke-a-text-encoder
clip comfyui comfyui-nodes embeddings flux1 remove sdxl t5 text-encoder
Last synced: 6 days ago
JSON representation
For SDXL, SD1.5, Flux. Nuke T5 and let CLIP guide Flux.1 on its own! Or let let random guide Flux.1! Or load a CLIP crazy opinion embedding about your image and let that guide the AI!
- Host: GitHub
- URL: https://github.com/zer0int/comfyui-nuke-a-text-encoder
- Owner: zer0int
- Created: 2024-12-09T01:16:54.000Z (14 days ago)
- Default Branch: CLIP-vision
- Last Pushed: 2024-12-09T01:30:58.000Z (14 days ago)
- Last Synced: 2024-12-09T02:26:39.940Z (14 days ago)
- Topics: clip, comfyui, comfyui-nodes, embeddings, flux1, remove, sdxl, t5, text-encoder
- Language: Python
- Homepage:
- Size: 19.5 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
### ComfyUI Nukes & Embeddings for Text Encoders! 🤯
## Update 09/DEC/24:
- Added random seed control for "randn" of Text Encoders (random, fixed, in-/decrement)
- ⚠️ After update, when using existing workflows, please RIGHT CLICK -> FIX NODE (rebuild)!
------
- Put the "ComfyUI-Nuke-a-TE" folder into "ComfyUI/custom_nodes" and run Comfy.
- Nuke a text encoder (zero the image-guiding input)!
- Nuke T5 to guide Flux.1-dev with CLIP only!
- (Make AI crazy again! 🤪)
- Use a random distribution (torch.randn) for CLIP *and* T5! 🥳
- Explore Flux.1's bias as it stares into itelf! 👀
- Nodes (and workflows) for Flux.1, SD-1.5, SDXL ✅
- You should totally [let CLIP rant about your image](https://github.com/zer0int/CLIP-gradient-ascent-embeddings) and use that embedding to steer an image. 🤗
- You should totally [download my new CLIP-SAE Text Encoder](https://huggingface.co/zer0int/CLIP-SAE-ViT-L-14) from HuggingFace! 🤗
- Node for SDXL is work of art by OpenAI's o1; the AI decided to slice the tensor and inject the CLIP embedding as it couldn't find where to separate it. This is very wrong, and very right, and totally an AI's art, and so be it! Slice those tensors and make it more... madness! 🤖
------
o1 did this (SDXL).
------
Flux.1 guided by a double random distribution bias reveal.
------
CLIP still can't write coherent text, and never will. But it makes fun things on its own! (Flux.1-dev)
------
My new CLIP fine-tune, linked above! (CLIP only Flux.1 guidance, 🚫no T5💥
------
A bit of love & strangeness for good old SD-1.5 🙃
