AbstractPhil commited on
Commit
73ec3a4
·
verified ·
1 Parent(s): c71a629

Model card: Anima non-commercial license (was the MIT placeholder), files, prompting, training

Browse files
Files changed (1) hide show
  1. README.md +91 -1
README.md CHANGED
@@ -1,3 +1,93 @@
1
  ---
2
- license: mit
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: other
3
+ license_name: circlestone-labs-non-commercial-license
4
+ license_link: https://huggingface.co/circlestone-labs/Anima/blob/main/LICENSE.md
5
+ base_model: circlestone-labs/Anima
6
+ base_model_relation: adapter
7
+ pipeline_tag: text-to-image
8
+ tags:
9
+ - lora
10
+ - anima
11
+ - comfyui
12
+ - liminal
13
+ - text-to-image
14
+ datasets:
15
+ - AbstractPhil/mega-liminal
16
  ---
17
+
18
+ # Mega Liminal LoRA
19
+
20
+ A LoRA for liminal spaces: empty malls, fog-bound roads, parking garages, suburbs at night, vacant theatres
21
+ and hallways. It was trained on the 1,873 curated and captioned images of the
22
+ [mega-liminal dataset](https://huggingface.co/datasets/AbstractPhil/mega-liminal).
23
+
24
+ > **NON-COMMERCIAL.** These files are derived from [Anima](https://huggingface.co/circlestone-labs/Anima)
25
+ > and inherit its license: the CircleStone Labs Non-Commercial License, plus the NVIDIA Open Model License
26
+ > of the Cosmos-Predict2 weights underneath it. No commercial use.
27
+
28
+ **Status:** training (started 2026-09-26 00:14 UTC). A new file lands every 2 epochs. Sample images come
29
+ when the run finishes.
30
+
31
+ ## Files
32
+
33
+ The folders are `<base model>/<LoRA size>/`.
34
+
35
+ | Path | What it is |
36
+ |---|---|
37
+ | `anima/rank64/mega-liminal-anima-r64-eNN.safetensors` | the LoRA after epoch NN (02, 04, ... 16; e16 is the final one), ComfyUI format |
38
+ | `anima/rank64/adapter_config.json` | the PEFT adapter config |
39
+ | `anima/rank64/training/` | the training configs and the TensorBoard log |
40
+
41
+ ## Using it in ComfyUI
42
+
43
+ 1. Get the Anima base files from [circlestone-labs/Anima](https://huggingface.co/circlestone-labs/Anima)
44
+ `split_files/`: `anima-base-v1.0.safetensors` (diffusion_models), `qwen_3_06b_base.safetensors`
45
+ (text_encoders) and `qwen_image_vae.safetensors` (vae).
46
+ 2. Put the LoRA file in `models/loras` and load it with **LoraLoaderModelOnly** at strength 1.0.
47
+ 3. Sample around 1 megapixel (1024x1024, or anything from 1:2 to 2:1) with 30-50 steps and CFG 4-5
48
+ (er_sde or euler_a), the settings the Anima card recommends for the base model.
49
+
50
+ ## Prompting
51
+
52
+ Every training caption has the same shape:
53
+
54
+ ```
55
+ liminal, <class>, <two to four plain sentences describing the image>
56
+ ```
57
+
58
+ The classes are: `mega liminal`, `landscape`, `suburban`, `simulacrum`, `vanishing point`, `empty mall`,
59
+ `hand sourced`, `parking garage`, `cityscape`, `movie theatre`.
60
+
61
+ The sentences name the kind of image, the place and its layout, the camera position, the architecture and
62
+ materials, and the lighting and colors. Prompts written the same way work best. Two captions from the
63
+ training set:
64
+
65
+ ```
66
+ liminal, mega liminal, A photograph of a snowy landscape dominated by dense fog that obscures most of the scene. In the middle ground, a blue road sign stands on two thin metal poles, its text illegible due to the heavy mist. The foreground and background merge into a uniform gray-white expanse with no visible terrain features, trees, or sky. Lighting is flat and diffused, typical of overcast winter weather, with no discernible shadows.
67
+ ```
68
+
69
+ ```
70
+ liminal, parking garage, Aerial photograph of a large, empty parking lot with neatly arranged rows of marked spaces. The asphalt surface is dark and smooth, marked with white lines and yellow curbs. Tall streetlights are evenly spaced throughout the lot. In the background, a modern building with a flat roof and large windows is visible.
71
+ ```
72
+
73
+ ## Training
74
+
75
+ | Setting | Value |
76
+ |---|---|
77
+ | Base model | Anima base v1.0 (the version the Anima card names for LoRA training) |
78
+ | Data | 1,873 images in 10 classes; small classes repeated up to 8x, so one epoch is 2,789 samples |
79
+ | Captions | written by Qwen3.5-9B from each image, one structured pass per image |
80
+ | Resolution | 1024, aspect-ratio buckets from 1:2 to 2:1 |
81
+ | LoRA | rank 64 |
82
+ | Optimizer | Adam, learning rate 2e-5, betas 0.9 / 0.99, no weight decay, 100 warmup steps, gradient clip 1.0 |
83
+ | Batch | 8 (4 per GPU on 2x RTX 5090) |
84
+ | Length | 16 epochs, about 5,580 steps |
85
+ | Trainer | [diffusion-pipe](https://github.com/AbstractEyes/diffusion-pipe) (AbstractEyes fork), DeepSpeed data parallel |
86
+
87
+ Images under 512x512 pixels, game screenshots, low-poly and cartoon renders and pixel art were removed
88
+ before training: 202 of the 2,075 source images.
89
+
90
+ ## Dataset
91
+
92
+ The images and captions are at [AbstractPhil/mega-liminal](https://huggingface.co/datasets/AbstractPhil/mega-liminal),
93
+ laid out for any trainer that reads image and `.txt` caption pairs. Train your own liminal LoRA on it.