Card: link the upstream autoencoder and text encoder
Browse files
README.md
CHANGED
|
@@ -16,6 +16,8 @@ learnable layer-slice mixer (10.0M), and the AE decoder retrained as a pixel-spa
|
|
| 16 |
|---|---|
|
| 17 |
| output | 256×256, sides multiple of 64 |
|
| 18 |
| sampler | 8–16 Euler steps, cfg 1.5–2.5, 4–8 head steps |
|
|
|
|
|
|
|
| 19 |
| weights | fp16, 803 MiB bundle |
|
| 20 |
| training | FLUX-Reason-6M, one RTX 4080 |
|
| 21 |
| licence | MIT |
|
|
@@ -63,5 +65,7 @@ python3 src/sample.py --prompt "A macro photograph of a blue morpho butterfly."
|
|
| 63 |
- Animal anatomy fails (see the fox), hands fail; detail finer than 8 px is invented by the decoder.
|
| 64 |
- Loose caption adherence, no glyphs, 256² only. Research artifact — no safety tuning, no filter.
|
| 65 |
|
| 66 |
-
MIT for our code and weights (`src/`, `transformer/`); bundled
|
| 67 |
-
|
|
|
|
|
|
|
|
|
| 16 |
|---|---|
|
| 17 |
| output | 256×256, sides multiple of 64 |
|
| 18 |
| sampler | 8–16 Euler steps, cfg 1.5–2.5, 4–8 head steps |
|
| 19 |
+
| latent | 8× spatial / 32 channels from [`Muinez/f8c32-ae`](https://huggingface.co/Muinez/f8c32-ae), bundled in `vae/` |
|
| 20 |
+
| text encoder | [`jhu-clsp/ettin-encoder-150m`](https://huggingface.co/jhu-clsp/ettin-encoder-150m), frozen, bundled in `text_encoder/` |
|
| 21 |
| weights | fp16, 803 MiB bundle |
|
| 22 |
| training | FLUX-Reason-6M, one RTX 4080 |
|
| 23 |
| licence | MIT |
|
|
|
|
| 65 |
- Animal anatomy fails (see the fox), hands fail; detail finer than 8 px is invented by the decoder.
|
| 66 |
- Loose caption adherence, no glyphs, 256² only. Research artifact — no safety tuning, no filter.
|
| 67 |
|
| 68 |
+
MIT for our code and weights (`src/`, `transformer/`); the bundled text encoder
|
| 69 |
+
[Ettin-150M](https://huggingface.co/jhu-clsp/ettin-encoder-150m) is MIT, the autoencoder
|
| 70 |
+
[f8c32-ae](https://huggingface.co/Muinez/f8c32-ae) is Apache-2.0. Architecture notes: `docs/ARCH.md`,
|
| 71 |
+
Russian slides: `docs/presentation.pdf`.
|