mlboydaisuke commited on
Commit
7a2eaed
·
verified ·
1 Parent(s): 2a4daec

Document multi-reference in-context editing

Browse files
Files changed (1) hide show
  1. README.md +10 -0
README.md CHANGED
@@ -41,6 +41,8 @@ Qwen3 text encoder.
41
  | `VAEEncoder.aimodel` | 1024×1024 RGB image → latent (image-to-image / editing) |
42
  | `Transformer_edit.aimodel` | In-context edit DiT — 1024, sequence 8192 (output + 1 reference) |
43
  | `Transformer_edit_512.aimodel` | In-context edit DiT — 512, sequence 2048 |
 
 
44
  | `tokenizer/`, `pipeline.json`, `vae_bn_*.npy` | Sidecar assets (auto-loaded) |
45
 
46
  Weights are 4-bit quantized (int4, per-block, block size 32); compute precision
@@ -95,6 +97,14 @@ The stock apple/coreai-models runtime does text- and image-to-image only.
95
 
96
  int4, ~25 s for a 1024 edit on a Mac GPU (4 steps, guidance 1.0).
97
 
 
 
 
 
 
 
 
 
98
  ## How it was converted
99
 
100
  ```bash
 
41
  | `VAEEncoder.aimodel` | 1024×1024 RGB image → latent (image-to-image / editing) |
42
  | `Transformer_edit.aimodel` | In-context edit DiT — 1024, sequence 8192 (output + 1 reference) |
43
  | `Transformer_edit_512.aimodel` | In-context edit DiT — 512, sequence 2048 |
44
+ | `Transformer_edit_2ref.aimodel` | Two-reference edit DiT — 1024, sequence 12288 (output + 2 references) |
45
+ | `Transformer_edit_2ref_512.aimodel` | Two-reference edit DiT — 512, sequence 3072 |
46
  | `tokenizer/`, `pipeline.json`, `vae_bn_*.npy` | Sidecar assets (auto-loaded) |
47
 
48
  Weights are 4-bit quantized (int4, per-block, block size 32); compute precision
 
97
 
98
  int4, ~25 s for a 1024 edit on a Mac GPU (4 steps, guidance 1.0).
99
 
100
+ ### Multi-reference
101
+
102
+ **`Transformer_edit_2ref.aimodel`** takes **two** reference images at once — each concatenated at
103
+ its own time index (`T=10`, `T=20`) — so the instruction can combine them: *"put the subject from
104
+ the first image into the scene from the second image."* Same mechanism, longer sequence (12288).
105
+ `editImages(referenceImages:)` selects the 1- or 2-reference transformer by the number of images.
106
+ int4, ~43 s for a 1024 two-reference edit on a Mac GPU.
107
+
108
  ## How it was converted
109
 
110
  ```bash