anima
decoder
sorryhyun commited on
Commit
ed9da68
·
verified ·
1 Parent(s): a26e48a

Replace the head with realr_1119_2200 (original-image target + LPIPS at the base points, turbo point on v4_longer); previous head stays in old/

Browse files
Files changed (2) hide show
  1. README.md +8 -2
  2. anima_pixel_head.safetensors +2 -2
README.md CHANGED
@@ -23,7 +23,7 @@ writes RGB pixels directly, in place of the VAE decode. The head and its trainin
23
  | Output | a 16×16 RGB patch per token, in [-1, 1] |
24
  | Architecture | linear projection of the feature + linear embedding of the x_t patch, 2 ConvNeXt, one 2× pixel-shuffle upsample, 1 ConvNeXt, linear to 8×8×3 |
25
  | DiT | `anima-base-v1.0` (28 blocks) |
26
- | Target | VAE(x̂0) of the same forward, x̂0 = x_t − σ·v |
27
 
28
 
29
  ## Compatibility
@@ -34,10 +34,16 @@ Read the head only where it was trained:
34
  Spectrum included (its last step is an actual forward).
35
  - **Turbo**: the 4-step turbo LoRA of
36
  [anima-turbo-4step](https://huggingface.co/sorryhyun/anima-turbo-4step), Euler 4 steps,
37
- shift 3, CFG 1 (last-step σ 0.5).
 
 
 
38
 
39
  ## Benchmark
40
 
 
 
 
41
  ### In ComfyUI
42
 
43
  Turbo 4 steps, CFG 1, RTX 5070 Ti, 16 prompts at ~1 MP, one seed
 
23
  | Output | a 16×16 RGB patch per token, in [-1, 1] |
24
  | Architecture | linear projection of the feature + linear embedding of the x_t patch, 2 ConvNeXt, one 2× pixel-shuffle upsample, 1 ConvNeXt, linear to 8×8×3 |
25
  | DiT | `anima-base-v1.0` (28 blocks) |
26
+ | Target | base points: the original image, with LPIPS (weight 0.05); turbo point: VAE(x̂0) of the same forward, x̂0 = x_t − σ·v |
27
 
28
 
29
  ## Compatibility
 
34
  Spectrum included (its last step is an actual forward).
35
  - **Turbo**: the 4-step turbo LoRA of
36
  [anima-turbo-4step](https://huggingface.co/sorryhyun/anima-turbo-4step), Euler 4 steps,
37
+ shift 3, CFG 1 (last-step σ 0.5). Trained on `anima_turbo_4step_v4_longer.safetensors`.
38
+
39
+ The previous head is `old/anima_pixel_head.safetensors` (VAE(x̂0) target at every point, turbo
40
+ point on that repo's `old/anima_turbo_4step.safetensors`).
41
 
42
  ## Benchmark
43
 
44
+ The benchmarks, samples and held-out reads below were measured on the previous head
45
+ (`old/anima_pixel_head.safetensors`).
46
+
47
  ### In ComfyUI
48
 
49
  Turbo 4 steps, CFG 1, RTX 5070 Ti, 16 prompts at ~1 MP, one seed
anima_pixel_head.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:27b3f0dc8297fa327bac636e1eeb962f1b93a3144445c08e5cc96d9e5d96b9d1
3
- size 99442152
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b9d24cfb84a38aa9c764f21ed47496f6dd2ccb2ceff063f467ed3d1a7f4a5a2
3
+ size 99442192