Zimeng Xiong
|
fe6936449d
|
Add Qwen-Image-Layered support for image decomposition into RGBA layers
This PR adds support for the Qwen-Image-Layered model, which decomposes an input
image into semantically disentangled RGBA layers for layer-based editing workflows.
## Features
- New CLI command: \`mflux-generate-qwen-layered\`
- Decomposes images into N RGBA layers (default 4)
- Supports 6-bit quantization for ~29GB memory usage (vs 55GB BF16)
- Resolution buckets: 640 and 1024
## Architecture
- RGBA-VAE (4-channel) with 3D temporal convolutions for layer handling
- Layer3D RoPE: 3D positional encoding [layer, height, width]
- Uses base QwenTransformer with extended RoPE for multi-layer sequences
- Condition image encoded as layer_index=-1 for proper decomposition
## New Files
- \`src/mflux/models/qwen_layered/\` - Full model implementation
- \`model/qwen_layered_vae/\` - RGBA-VAE encoder/decoder
- \`model/qwen_layered_transformer/\` - Layer3D RoPE
- \`weights/\` - Weight mapping and definitions
- \`variants/i2l/\` - Image-to-Layers pipeline
- \`cli/\` - Command-line interface
## Usage
\`\`\`sh
mflux-generate-qwen-layered \\
--image input.png \\
--layers 4 \\
--steps 50 \\
-q 6 \\
--output-dir ./layers
\`\`\`
## Documentation
Added comprehensive documentation to README.md including:
- TOC entry
- CLI argument reference
- Usage examples and tips
- Memory requirements
Tested on M4 Max 48GB with 6-bit quantization.
|
2025-12-22 08:08:20 -08:00 |
|
Filip Strand
|
f45082f880
|
Small LoRA improvements (#291)
|
2025-12-05 03:18:27 +01:00 |
|
filipstrand
|
5369d5adc6
|
Update VLM prompt to match
|
2025-12-04 01:49:55 +01:00 |
|
Filip Strand
|
8e5609132c
|
Release 0.13.0 (#289)
|
2025-12-03 22:40:44 +01:00 |
|
Filip Strand
|
6551e9f6a1
|
Update dependencies (#288)
|
2025-12-03 22:13:55 +01:00 |
|
Filip Strand
|
37d202d0df
|
Add support for Z-image Turbo & major improvements for weight loading and parameter resolution (#284)
|
2025-12-03 15:19:32 +01:00 |
|
Filip Strand
|
eb83175680
|
Fix: Auto-download tokenizer files for FIBO-vlm when missing (#283)
|
2025-11-27 18:15:56 +01:00 |
|
Filip Strand
|
c2c860f4de
|
Add Bria FIBO support (#279)
Co-authored-by: Filip Strand <filip@host-022.local>
|
2025-11-27 13:01:55 +01:00 |
|
Filip Strand
|
2e92984125
|
Release 0.11.1 (#278)
Co-authored-by: Filip Strand <filip@Host-022.local>
|
2025-11-13 01:05:29 +01:00 |
|
Filip Strand
|
d314aa6889
|
Release 0.11.0 (#271)
|
2025-10-14 00:34:27 +02:00 |
|
Anthony Wu
|
6947a23d6c
|
Scheduler interface implementation. Implement pre-existing linear scheduler only. (#258)
Co-authored-by: Anthony Wu <pls-file-gh-issue@users.noreply.github.com>
Co-authored-by: filipstrand <strand.filip@gmail.com>
|
2025-10-12 12:20:40 +02:00 |
|
Filip Strand
|
98e03a46f2
|
Add Qwen Image support (#269)
|
2025-10-06 07:41:14 +02:00 |
|
Anthony Wu
|
0d9af4f380
|
Python 3.13 sentencepiece and torch unblocked (#259)
Co-authored-by: Anthony Wu <pls-file-gh-issue@users.noreply.github.com>
|
2025-08-31 22:06:26 +02:00 |
|
Filip Strand
|
beee9b9509
|
Add support for FLUX.1-Krea-dev model (#241)
|
2025-08-03 00:41:00 +02:00 |
|
filipstrand
|
df66510380
|
Sync lockfile
|
2025-07-20 12:13:12 +02:00 |
|
Filip Strand
|
3c5df450cc
|
Release v.0.9.4 (#237)
|
2025-07-17 11:12:53 +02:00 |
|