Phil2Sat commited on
Commit
82fc3c2
·
1 Parent(s): 80137f1
Files changed (1) hide show
  1. README.md +10 -18
README.md CHANGED
@@ -12,46 +12,38 @@ tags:
12
  - t2i
13
  - i2i
14
  ---
15
-
16
- Only quants of https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO.
17
 
18
  READ CAREFULLY!!!
19
 
20
- I think i have fixed the smaller quants, feel free to test and give feedback.
21
 
22
- I'll leave the warning anyway.
23
 
24
- ATTENTION !!! the smaller quants, smaller than Q5 except Q4_K_M creates weird artifacts, see:
25
- https://huggingface.co/QuantStack/Qwen-Image-Edit-2509-GGUF/discussions/6#68e7099a87b5c3907529d4bc
26
 
27
 
28
- no more AIO, needs text_encoder and vae
29
 
30
 
31
- personally i recommend:
32
 
33
  text_encoder https://huggingface.co/mradermacher/Qwen2.5-VL-7B-Instruct-abliterated-GGUF/tree/main
34
-
35
  mmproj needs clip fix in ComfyUI-GGUF/loader.py line 13
36
  ```
37
  TXT_ARCH_LIST = {"t5", "t5encoder", "llama", "qwen2vl", "clip"}
38
  ```
39
  see https://github.com/city96/ComfyUI-GGUF/pull/349
40
-
41
- or i have cloned the abliberated files into my repo and fixed the mmroj so the fix is no longer needed
42
-
43
-
44
  VAE https://huggingface.co/calcuis/pig-vae/blob/main/pig_qwen_image_vae_fp32-f16.gguf
45
-
46
-
47
- replace comfy_extras/nodes_qwen.py with https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/blob/main/fixed-textencode-node/nodes_qwen.py
48
 
49
 
50
  workflow text2image 4-step Q2_K:
51
  ![ComfyUI_00574_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/0ciWa9SH5VRfdk8PdZCWr.png)
52
-
53
  workflow edit of this image 4-step Q2_K:
54
  ![ComfyUI_00601_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/o0-AjNdwAJmTzX8r-ncVs.png)
55
 
56
  workflow edit of this image 4-step Q5_K_M:
57
- ![ComfyUI_00602_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/skXvWdUiWFJ9m8L3WSs-C.png)
 
12
  - t2i
13
  - i2i
14
  ---
15
+ Only quants of https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO
 
16
 
17
  READ CAREFULLY!!!
18
 
19
+ I think I have fixed the smaller quants, feel free to test and give a feedback.
20
 
21
+ I left the warning anyway.
22
 
23
+ ATTENTION !!! The smaller quants, smaller than Q5 except Q4_K_M creates weird artifacts,
24
+ see https://huggingface.co/QuantStack/Qwen-Image-Edit-2509-GGUF/discussions/6#68e7099a87b5c3907529d4bc
25
 
26
 
27
+ Not AIO, needs the text encoder and the VAE
28
 
29
 
30
+ I personally recommend:
31
 
32
  text_encoder https://huggingface.co/mradermacher/Qwen2.5-VL-7B-Instruct-abliterated-GGUF/tree/main
 
33
  mmproj needs clip fix in ComfyUI-GGUF/loader.py line 13
34
  ```
35
  TXT_ARCH_LIST = {"t5", "t5encoder", "llama", "qwen2vl", "clip"}
36
  ```
37
  see https://github.com/city96/ComfyUI-GGUF/pull/349
38
+ I have cloned the abliberated files into my repo and fixed the mmproj, so the fix is no longer needed
 
 
 
39
  VAE https://huggingface.co/calcuis/pig-vae/blob/main/pig_qwen_image_vae_fp32-f16.gguf
40
+ Replace comfy_extras/nodes_qwen.py with https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/blob/main/fixed-textencode-node/nodes_qwen.py
 
 
41
 
42
 
43
  workflow text2image 4-step Q2_K:
44
  ![ComfyUI_00574_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/0ciWa9SH5VRfdk8PdZCWr.png)
 
45
  workflow edit of this image 4-step Q2_K:
46
  ![ComfyUI_00601_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/o0-AjNdwAJmTzX8r-ncVs.png)
47
 
48
  workflow edit of this image 4-step Q5_K_M:
49
+ ![ComfyUI_00602_](https://cdn-uploads.huggingface.co/production/uploads/650eac15f26dbb8827b1d443/skXvWdUiWFJ9m8L3WSs-C.png)