devin-lai's picture
Prepare inference-only v0.5.0 runtime cleanup
63e9703 verified
|
Raw History Blame
5.38 kB
# Recorded sample images
These are authentic **AI-generated** outputs from recorded local MLX runs.
The original PNGs, inference records, editing input and comparison grids remain
in the [access-reviewed Hugging Face gallery](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples)
under the upstream model terms. The public catalog retains exact prompts,
seeds, dimensions and PNG SHA-256 identities; no sample has been retouched.
Run a sample after installing the runtime and downloading packed12:
```bash
python examples/generate.py --model models/packed12 \
--text-encoder models/packed12/text_encoder \
--case product --output outputs/product.png
```
Choose `portrait` or `holdout-detail` in the same way. For `edit`, download
the [original reference](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/edit-input.png)
after access approval and pass `--image /path/to/edit-input.png`.
Using another input produces a different experiment. All examples use the
six original heads, PyTorch CPU noise, packed12 and the measured exact tuning
profile with compiled VAE batch four. Byte identity depends on the tested
hardware/software and input files. [Machine-readable catalog](catalog.json).
## product
[View the recorded image](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/product.png) 路 [Full inference JSON](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/product.json)
Seed `731` 路 output `2048 脳 2048` 路 generation canvas `2048 脳 2048`.
```text
Square editorial studio photograph of a dark cobalt glass perfume bottle on pale travertine. The centered cream label reads exactly 'NIGHT TIDE' in large engraved serif letters and 'EAU DE PARFUM' in small capitals beneath. Dappled blue caustics, crisp glass edges, fine pores in the stone, dark reflections through the translucent bottle, restrained palette and generous negative space. No other text.
```
PNG SHA-256: `63684ef8ae148a35ddef02650c6aa30b4843b47c99b06cec2ea3708ef3fc2293`.
## portrait
[View the recorded image](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/portrait.png) 路 [Full inference JSON](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/portrait.json)
Seed `1234` 路 output `2048 脳 2048` 路 generation canvas `2048 脳 2048`.
```text
An intimate natural-light portrait of an elderly woman in a botanical conservatory, reading a small red clothbound book. Warm brown skin with visible freckles and fine wrinkles, silver curls lit from behind, intricate indigo knit cardigan. Soft fern fronds out of focus in the foreground; detailed hands and realistic skin texture. A candid 85mm color photograph with gentle shadows and no lettering.
```
PNG SHA-256: `32af5e3b95919e79e0019058ec396824c5861f9a43a4fa50c7b20348e8d81933`.
## holdout-detail
[View the recorded image](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/holdout-detail.png) 路 [Full inference JSON](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/holdout-detail.json)
Seed `928417` 路 output `2048 脳 2048` 路 generation canvas `2048 脳 2048`.
```text
A meticulous natural light still life on a wooden watchmaker's bench: a silver mechanical wristwatch with a clear glass back, brass gears and hair-thin springs, a folded cream linen glove, a small matte ceramic bowl and a card reading 'SECOND LIGHT'. In the background a woman with warm brown freckled skin holds a magnifying lens, her hands realistically detailed. Fine brushed metal, subtle skin pores, smooth dark-to-light blue gradients and soft window reflections, editorial macro photography.
```
PNG SHA-256: `ddbfc06c4556c08669d35a4618d380a99998e39a0226e7688d149993ab183038`.
## edit
[View the recorded image](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/edit.png) 路 [Full inference JSON](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/examples/edit.json)
Seed `42` 路 output `1024 脳 1024` 路 generation canvas `2048 脳 2048`.
```text
Change the cobalt blue glass perfume bottle in <Picture 1> to deep emerald green glass. Preserve the pale engraved NIGHT TIDE label, exact lettering, bottle shape, cap, stone backdrop, composition, lighting and shadows. The glass should be visibly green with realistic emerald highlights and reflections.
```
PNG SHA-256: `9df3f80efa5ef9f078c9acda81febdbc2d5d706f3c85c574fdfd9fd274b71304`.
## Compare weight choices
- [Native / Q8 / Q4 full-frame stack](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/comparisons/product/stack.png)
- [Matched native-resolution lettering/detail crops](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/comparisons/product/details.png)
- [Portrait comparison](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/comparisons/portrait/stack.png)
- [Original reference and emerald-glass edit](https://huggingface.co/devin-lai/MiniMax-H3-Image-mlx-examples/blob/v0.4.0/comparisons/edit-example/before-after.png)
Q8/Q4 change fine detail and lettering. Similarity scores are evidence for
these recorded cases, rather than a broad quality ranking.
[Measurement details](../BENCHMARKS.md).