Link visual examples from the model card

README.md

@@ -12,18 +12,80 @@ tags:
 
 # MiniMax-H3 Level-of-Token research release
 
-Experimental LoT adapter and rank-16 DiT LoRA, plus the full benchmark summary, per-source confirmation metrics, and matched example videos.
+Experimental LoT adapter and rank-16 DiT LoRA, with benchmarks and visual examples.
 
-**Research status:** the latest campaign found no branch that met the predeclared promotion threshold. Metrics show tradeoffs, not a confirmed quality win. See [`benchmarks/README.md`](benchmarks/README.md) and [`benchmarks/campaign-r4-results.json`](benchmarks/campaign-r4-results.json).
+**Research status:** the latest campaign found no branch that met the predeclared promotion threshold. Metrics show tradeoffs, not a confirmed quality win. See [benchmark results](benchmarks/README.md) and [campaign summary](benchmarks/campaign-r4-results.json).
 
-## Weight files
+## Weights
 
-- `adapter.safetensors`: LoT adapter and fitted extent bank.
-- `lora.safetensors`: rank-16 DiT LoRA.
+- [`adapter.safetensors`](adapter.safetensors): LoT adapter and fitted extent bank.
+- [`lora.safetensors`](lora.safetensors): rank-16 DiT LoRA.
 
-## Examples
+## Video comparison grids
 
-`examples/control/` and `examples/preserve/` contain the 30-step confirmation comparison grids, generated clips, and sanitized manifests for four held-out prompts per branch. The video files include dense-LoRA, detail50 warm-up, and uniform50 outputs; the comparison sheets also show source-reference panels.
+Both grids show matched 30-step outputs from the untouched confirmation prompts. The displayed comparison includes source-reference panels; the downloadable MP4 links below are generated outputs.
+
+### Control branch, step 500
+
+![Control branch comparison grid](examples/control/comparison.png)
+
+### Preserve branch, step 500
+
+![Preserve branch comparison grid](examples/preserve/comparison.png)
+
+## Generated video examples
+
+Each link opens an MP4 for one confirmation prompt. Prompts and seeds are in the branch [manifests](examples/control/manifest.json) and [preserve manifest](examples/preserve/manifest.json).
+
+### Control branch
+
+| Prompt | dense-LoRA | detail 50% | uniform 50% |
+|---|---|---|---|
+| p0 | [MP4](examples/control/p0_dense_lora.mp4) | [MP4](examples/control/p0_detail50_warmup.mp4) | [MP4](examples/control/p0_uniform50_1x2.mp4) |
+| p1 | [MP4](examples/control/p1_dense_lora.mp4) | [MP4](examples/control/p1_detail50_warmup.mp4) | [MP4](examples/control/p1_uniform50_1x2.mp4) |
+| p2 | [MP4](examples/control/p2_dense_lora.mp4) | [MP4](examples/control/p2_detail50_warmup.mp4) | [MP4](examples/control/p2_uniform50_1x2.mp4) |
+| p3 | [MP4](examples/control/p3_dense_lora.mp4) | [MP4](examples/control/p3_detail50_warmup.mp4) | [MP4](examples/control/p3_uniform50_1x2.mp4) |
+
+### Preserve branch
+
+| Prompt | dense-LoRA | detail 50% | uniform 50% |
+|---|---|---|---|
+| p0 | [MP4](examples/preserve/p0_dense_lora.mp4) | [MP4](examples/preserve/p0_detail50_warmup.mp4) | [MP4](examples/preserve/p0_uniform50_1x2.mp4) |
+| p1 | [MP4](examples/preserve/p1_dense_lora.mp4) | [MP4](examples/preserve/p1_detail50_warmup.mp4) | [MP4](examples/preserve/p1_uniform50_1x2.mp4) |
+| p2 | [MP4](examples/preserve/p2_dense_lora.mp4) | [MP4](examples/preserve/p2_detail50_warmup.mp4) | [MP4](examples/preserve/p2_uniform50_1x2.mp4) |
+| p3 | [MP4](examples/preserve/p3_dense_lora.mp4) | [MP4](examples/preserve/p3_detail50_warmup.mp4) | [MP4](examples/preserve/p3_uniform50_1x2.mp4) |
+
+## Still-image examples
+
+Earlier H3 still-image experiments are shown below; these are separate from the r4 video campaign.
+
+### Evaluation curve
+
+![Evaluation curve](examples/stills/lot-eval-curve.png)
+
+### H3 render — bands
+
+![H3 render — bands](examples/stills/lot-h3-render-bands.png)
+
+### H3 render — uniform 2×2
+
+![H3 render — uniform 2×2](examples/stills/lot-h3-render-uniform2.png)
+
+### Trained LoT — bands
+
+![Trained LoT — bands](examples/stills/lot-h3-trained-bands.png)
+
+### Trained LoT — uniform 2×2
+
+![Trained LoT — uniform 2×2](examples/stills/lot-h3-trained-uniform2.png)
+
+### Layout comparison
+
+![Layout comparison](examples/stills/lot-layout-compare.png)
+
+### Mixed-token grid
+
+![Mixed-token grid](examples/stills/lot-mixed-grid.png)
 
 ## Attribution