Update repo paths to AnimeLoom org namespace
Browse files
README.md
CHANGED
|
@@ -1,158 +1,158 @@
|
|
| 1 |
-
---
|
| 2 |
-
base_model: cagliostrolab/animagine-xl-3.1
|
| 3 |
-
library_name: peft
|
| 4 |
-
license: openrail++
|
| 5 |
-
pipeline_tag: text-to-image
|
| 6 |
-
tags:
|
| 7 |
-
- stable-diffusion-xl
|
| 8 |
-
- stable-diffusion
|
| 9 |
-
- lora
|
| 10 |
-
- peft
|
| 11 |
-
- anime
|
| 12 |
-
- character-lora
|
| 13 |
-
- animeloom
|
| 14 |
-
- denji
|
| 15 |
-
- chainsaw-man
|
| 16 |
-
instance_prompt: "1boy, denji, chainsaw man, blonde hair, yellow eyes"
|
| 17 |
-
widget:
|
| 18 |
-
- text: "1boy, denji, chainsaw man, blonde hair, yellow eyes, anime, masterpiece, best quality"
|
| 19 |
-
---
|
| 20 |
-
|
| 21 |
-
# AnimeLoom — Denji LoRA
|
| 22 |
-
|
| 23 |
-
Character LoRA for **Denji** (Chainsaw Man), trained as part of the
|
| 24 |
-
[AnimeLoom](https://github.com/JoelJohnsonThomas/AnimeLoom) anime
|
| 25 |
-
character-consistency pipeline.
|
| 26 |
-
|
| 27 |
-
Two adapters are provided in this repo:
|
| 28 |
-
|
| 29 |
-
| Folder | Base model | Rank | Steps | Images | Resolution |
|
| 30 |
-
|---------|----------------------------------|------|-------|--------|------------|
|
| 31 |
-
| `sdxl/` | `cagliostrolab/animagine-xl-3.1` | 32 | 2200 | 15 | 1024 |
|
| 32 |
-
| `sd15/` | `Lykon/dreamshaper-8` | 32 | 800 | 15 | 512 |
|
| 33 |
-
|
| 34 |
-
The SDXL adapter is the primary one used by AnimeLoom's identity-keyframe stage
|
| 35 |
-
(SDXL + character LoRA + IP-Adapter). The SD 1.5 adapter is provided for
|
| 36 |
-
compatibility with lighter inference pipelines (e.g. AnimateDiff workflows).
|
| 37 |
-
|
| 38 |
-
## Trigger words
|
| 39 |
-
|
| 40 |
-
```
|
| 41 |
-
1boy, denji, chainsaw man, blonde hair, yellow eyes
|
| 42 |
-
```
|
| 43 |
-
|
| 44 |
-
Add booru-style descriptors as needed (e.g. `black suit`, `chainsaw`,
|
| 45 |
-
`shark teeth`, `smiling`).
|
| 46 |
-
|
| 47 |
-
## Usage — SDXL with PEFT / diffusers
|
| 48 |
-
|
| 49 |
-
```python
|
| 50 |
-
import torch
|
| 51 |
-
from diffusers import StableDiffusionXLPipeline
|
| 52 |
-
from peft import PeftModel
|
| 53 |
-
|
| 54 |
-
pipe = StableDiffusionXLPipeline.from_pretrained(
|
| 55 |
-
"cagliostrolab/animagine-xl-3.1",
|
| 56 |
-
torch_dtype=torch.float16,
|
| 57 |
-
).to("cuda")
|
| 58 |
-
|
| 59 |
-
# AnimeLoom's training output is a PEFT adapter — load via PeftModel
|
| 60 |
-
pipe.unet = PeftModel.from_pretrained(
|
| 61 |
-
pipe.unet,
|
| 62 |
-
"
|
| 63 |
-
subfolder="sdxl",
|
| 64 |
-
)
|
| 65 |
-
|
| 66 |
-
img = pipe(
|
| 67 |
-
"1boy, denji, chainsaw man, blonde hair, yellow eyes, anime, "
|
| 68 |
-
"masterpiece, best quality, absurdres",
|
| 69 |
-
negative_prompt="blurry, low quality, deformed, extra fingers, 3d render",
|
| 70 |
-
num_inference_steps=28,
|
| 71 |
-
guidance_scale=6.5,
|
| 72 |
-
height=1024, width=1024,
|
| 73 |
-
).images[0]
|
| 74 |
-
img.save("denji.png")
|
| 75 |
-
```
|
| 76 |
-
|
| 77 |
-
## Usage — SD 1.5 with diffusers
|
| 78 |
-
|
| 79 |
-
```python
|
| 80 |
-
import torch
|
| 81 |
-
from diffusers import StableDiffusionPipeline
|
| 82 |
-
|
| 83 |
-
pipe = StableDiffusionPipeline.from_pretrained(
|
| 84 |
-
"Lykon/dreamshaper-8",
|
| 85 |
-
torch_dtype=torch.float16,
|
| 86 |
-
).to("cuda")
|
| 87 |
-
|
| 88 |
-
pipe.load_lora_weights(
|
| 89 |
-
"
|
| 90 |
-
subfolder="sd15",
|
| 91 |
-
weight_name="pytorch_lora_weights.safetensors",
|
| 92 |
-
)
|
| 93 |
-
|
| 94 |
-
img = pipe(
|
| 95 |
-
"1boy, denji, chainsaw man, blonde hair, anime, masterpiece",
|
| 96 |
-
num_inference_steps=28,
|
| 97 |
-
guidance_scale=7.0,
|
| 98 |
-
).images[0]
|
| 99 |
-
```
|
| 100 |
-
|
| 101 |
-
## Recommended weights & prompting
|
| 102 |
-
|
| 103 |
-
- **LoRA scale**: `0.85 - 1.15` (SDXL), `0.6 - 0.8` (SD 1.5)
|
| 104 |
-
- Pair with anime base models: Animagine XL 3.1, Counterfeit-V3.0, AnythingV5
|
| 105 |
-
- For face consistency in video, combine with **IP-Adapter SDXL** and
|
| 106 |
-
**Wan2.2-Animate** face-lock — see the AnimeLoom pipeline.
|
| 107 |
-
|
| 108 |
-
## AnimeLoom video pipeline integration
|
| 109 |
-
|
| 110 |
-
This LoRA is built to feed AnimeLoom's text-to-anime-video pipeline:
|
| 111 |
-
|
| 112 |
-
```
|
| 113 |
-
SDXL + this LoRA + IP-Adapter → identity keyframe (Phase 2)
|
| 114 |
-
↓
|
| 115 |
-
Wan2.2 I2V → motion driving clip (Phase 3a)
|
| 116 |
-
↓
|
| 117 |
-
Wan2.2-Animate face-lock → face from keyframe pasted onto motion (Phase 3b)
|
| 118 |
-
↓
|
| 119 |
-
RIFE + Real-ESRGAN + GFPGAN → temporal/spatial upscale + face restore (Phase 4)
|
| 120 |
-
↓
|
| 121 |
-
Final 24fps anime video with consistent character identity across shots.
|
| 122 |
-
```
|
| 123 |
-
|
| 124 |
-
## Limitations
|
| 125 |
-
|
| 126 |
-
- **Anime-only.** Photoreal prompts will degrade quality.
|
| 127 |
-
- **InsightFace face-swap does not work on these outputs** (it is photoreal-only).
|
| 128 |
-
For identity rescue use IP-Adapter-FaceID-SDXL or
|
| 129 |
-
[CharacterFaceSwap](https://github.com/ArtBot2023/CharacterFaceSwap).
|
| 130 |
-
- Trained on 15 images; expect occasional drift on unusual angles, complex
|
| 131 |
-
outfits, or transformation/hybrid forms (chainsaw devil hybrid).
|
| 132 |
-
- The SD 1.5 adapter is intentionally lower-fidelity (800 steps at 512 res)
|
| 133 |
-
for use as a lightweight fallback.
|
| 134 |
-
|
| 135 |
-
## License
|
| 136 |
-
|
| 137 |
-
Released under [OpenRAIL++](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
|
| 138 |
-
(`openrail++`), inheriting from the base model
|
| 139 |
-
[Animagine XL 3.1](https://huggingface.co/cagliostrolab/animagine-xl-3.1).
|
| 140 |
-
|
| 141 |
-
## Related models
|
| 142 |
-
|
| 143 |
-
Part of the AnimeLoom character collection:
|
| 144 |
-
|
| 145 |
-
- [
|
| 146 |
-
- [
|
| 147 |
-
- [
|
| 148 |
-
|
| 149 |
-
## Citation
|
| 150 |
-
|
| 151 |
-
If you use this LoRA in research or production, please credit AnimeLoom and
|
| 152 |
-
the base model authors (Cagliostro Lab for Animagine XL 3.1, Lykon for
|
| 153 |
-
DreamShaper 8).
|
| 154 |
-
|
| 155 |
-
---
|
| 156 |
-
|
| 157 |
-
*Trained with [PEFT](https://huggingface.co/docs/peft) as part of the
|
| 158 |
-
[AnimeLoom](https://github.com/JoelJohnsonThomas/AnimeLoom) project.*
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: cagliostrolab/animagine-xl-3.1
|
| 3 |
+
library_name: peft
|
| 4 |
+
license: openrail++
|
| 5 |
+
pipeline_tag: text-to-image
|
| 6 |
+
tags:
|
| 7 |
+
- stable-diffusion-xl
|
| 8 |
+
- stable-diffusion
|
| 9 |
+
- lora
|
| 10 |
+
- peft
|
| 11 |
+
- anime
|
| 12 |
+
- character-lora
|
| 13 |
+
- animeloom
|
| 14 |
+
- denji
|
| 15 |
+
- chainsaw-man
|
| 16 |
+
instance_prompt: "1boy, denji, chainsaw man, blonde hair, yellow eyes"
|
| 17 |
+
widget:
|
| 18 |
+
- text: "1boy, denji, chainsaw man, blonde hair, yellow eyes, anime, masterpiece, best quality"
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
# AnimeLoom — Denji LoRA
|
| 22 |
+
|
| 23 |
+
Character LoRA for **Denji** (Chainsaw Man), trained as part of the
|
| 24 |
+
[AnimeLoom](https://github.com/JoelJohnsonThomas/AnimeLoom) anime
|
| 25 |
+
character-consistency pipeline.
|
| 26 |
+
|
| 27 |
+
Two adapters are provided in this repo:
|
| 28 |
+
|
| 29 |
+
| Folder | Base model | Rank | Steps | Images | Resolution |
|
| 30 |
+
|---------|----------------------------------|------|-------|--------|------------|
|
| 31 |
+
| `sdxl/` | `cagliostrolab/animagine-xl-3.1` | 32 | 2200 | 15 | 1024 |
|
| 32 |
+
| `sd15/` | `Lykon/dreamshaper-8` | 32 | 800 | 15 | 512 |
|
| 33 |
+
|
| 34 |
+
The SDXL adapter is the primary one used by AnimeLoom's identity-keyframe stage
|
| 35 |
+
(SDXL + character LoRA + IP-Adapter). The SD 1.5 adapter is provided for
|
| 36 |
+
compatibility with lighter inference pipelines (e.g. AnimateDiff workflows).
|
| 37 |
+
|
| 38 |
+
## Trigger words
|
| 39 |
+
|
| 40 |
+
```
|
| 41 |
+
1boy, denji, chainsaw man, blonde hair, yellow eyes
|
| 42 |
+
```
|
| 43 |
+
|
| 44 |
+
Add booru-style descriptors as needed (e.g. `black suit`, `chainsaw`,
|
| 45 |
+
`shark teeth`, `smiling`).
|
| 46 |
+
|
| 47 |
+
## Usage — SDXL with PEFT / diffusers
|
| 48 |
+
|
| 49 |
+
```python
|
| 50 |
+
import torch
|
| 51 |
+
from diffusers import StableDiffusionXLPipeline
|
| 52 |
+
from peft import PeftModel
|
| 53 |
+
|
| 54 |
+
pipe = StableDiffusionXLPipeline.from_pretrained(
|
| 55 |
+
"cagliostrolab/animagine-xl-3.1",
|
| 56 |
+
torch_dtype=torch.float16,
|
| 57 |
+
).to("cuda")
|
| 58 |
+
|
| 59 |
+
# AnimeLoom's training output is a PEFT adapter — load via PeftModel
|
| 60 |
+
pipe.unet = PeftModel.from_pretrained(
|
| 61 |
+
pipe.unet,
|
| 62 |
+
"AnimeLoom/denji",
|
| 63 |
+
subfolder="sdxl",
|
| 64 |
+
)
|
| 65 |
+
|
| 66 |
+
img = pipe(
|
| 67 |
+
"1boy, denji, chainsaw man, blonde hair, yellow eyes, anime, "
|
| 68 |
+
"masterpiece, best quality, absurdres",
|
| 69 |
+
negative_prompt="blurry, low quality, deformed, extra fingers, 3d render",
|
| 70 |
+
num_inference_steps=28,
|
| 71 |
+
guidance_scale=6.5,
|
| 72 |
+
height=1024, width=1024,
|
| 73 |
+
).images[0]
|
| 74 |
+
img.save("denji.png")
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
## Usage — SD 1.5 with diffusers
|
| 78 |
+
|
| 79 |
+
```python
|
| 80 |
+
import torch
|
| 81 |
+
from diffusers import StableDiffusionPipeline
|
| 82 |
+
|
| 83 |
+
pipe = StableDiffusionPipeline.from_pretrained(
|
| 84 |
+
"Lykon/dreamshaper-8",
|
| 85 |
+
torch_dtype=torch.float16,
|
| 86 |
+
).to("cuda")
|
| 87 |
+
|
| 88 |
+
pipe.load_lora_weights(
|
| 89 |
+
"AnimeLoom/denji",
|
| 90 |
+
subfolder="sd15",
|
| 91 |
+
weight_name="pytorch_lora_weights.safetensors",
|
| 92 |
+
)
|
| 93 |
+
|
| 94 |
+
img = pipe(
|
| 95 |
+
"1boy, denji, chainsaw man, blonde hair, anime, masterpiece",
|
| 96 |
+
num_inference_steps=28,
|
| 97 |
+
guidance_scale=7.0,
|
| 98 |
+
).images[0]
|
| 99 |
+
```
|
| 100 |
+
|
| 101 |
+
## Recommended weights & prompting
|
| 102 |
+
|
| 103 |
+
- **LoRA scale**: `0.85 - 1.15` (SDXL), `0.6 - 0.8` (SD 1.5)
|
| 104 |
+
- Pair with anime base models: Animagine XL 3.1, Counterfeit-V3.0, AnythingV5
|
| 105 |
+
- For face consistency in video, combine with **IP-Adapter SDXL** and
|
| 106 |
+
**Wan2.2-Animate** face-lock — see the AnimeLoom pipeline.
|
| 107 |
+
|
| 108 |
+
## AnimeLoom video pipeline integration
|
| 109 |
+
|
| 110 |
+
This LoRA is built to feed AnimeLoom's text-to-anime-video pipeline:
|
| 111 |
+
|
| 112 |
+
```
|
| 113 |
+
SDXL + this LoRA + IP-Adapter → identity keyframe (Phase 2)
|
| 114 |
+
↓
|
| 115 |
+
Wan2.2 I2V → motion driving clip (Phase 3a)
|
| 116 |
+
↓
|
| 117 |
+
Wan2.2-Animate face-lock → face from keyframe pasted onto motion (Phase 3b)
|
| 118 |
+
↓
|
| 119 |
+
RIFE + Real-ESRGAN + GFPGAN → temporal/spatial upscale + face restore (Phase 4)
|
| 120 |
+
↓
|
| 121 |
+
Final 24fps anime video with consistent character identity across shots.
|
| 122 |
+
```
|
| 123 |
+
|
| 124 |
+
## Limitations
|
| 125 |
+
|
| 126 |
+
- **Anime-only.** Photoreal prompts will degrade quality.
|
| 127 |
+
- **InsightFace face-swap does not work on these outputs** (it is photoreal-only).
|
| 128 |
+
For identity rescue use IP-Adapter-FaceID-SDXL or
|
| 129 |
+
[CharacterFaceSwap](https://github.com/ArtBot2023/CharacterFaceSwap).
|
| 130 |
+
- Trained on 15 images; expect occasional drift on unusual angles, complex
|
| 131 |
+
outfits, or transformation/hybrid forms (chainsaw devil hybrid).
|
| 132 |
+
- The SD 1.5 adapter is intentionally lower-fidelity (800 steps at 512 res)
|
| 133 |
+
for use as a lightweight fallback.
|
| 134 |
+
|
| 135 |
+
## License
|
| 136 |
+
|
| 137 |
+
Released under [OpenRAIL++](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
|
| 138 |
+
(`openrail++`), inheriting from the base model
|
| 139 |
+
[Animagine XL 3.1](https://huggingface.co/cagliostrolab/animagine-xl-3.1).
|
| 140 |
+
|
| 141 |
+
## Related models
|
| 142 |
+
|
| 143 |
+
Part of the [AnimeLoom](https://huggingface.co/AnimeLoom) character collection:
|
| 144 |
+
|
| 145 |
+
- [AnimeLoom/sakura-haruno](https://huggingface.co/AnimeLoom/sakura-haruno)
|
| 146 |
+
- [AnimeLoom/denji](https://huggingface.co/AnimeLoom/denji) (this card)
|
| 147 |
+
- [AnimeLoom/yuki-nagato](https://huggingface.co/AnimeLoom/yuki-nagato)
|
| 148 |
+
|
| 149 |
+
## Citation
|
| 150 |
+
|
| 151 |
+
If you use this LoRA in research or production, please credit AnimeLoom and
|
| 152 |
+
the base model authors (Cagliostro Lab for Animagine XL 3.1, Lykon for
|
| 153 |
+
DreamShaper 8).
|
| 154 |
+
|
| 155 |
+
---
|
| 156 |
+
|
| 157 |
+
*Trained with [PEFT](https://huggingface.co/docs/peft) as part of the
|
| 158 |
+
[AnimeLoom](https://github.com/JoelJohnsonThomas/AnimeLoom) project.*
|