Upload folder using huggingface_hub

.gitattributes

@@ -35,3 +35,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 samples/v3_768x448_16steps_no-turbo.mp4 filter=lfs diff=lfs merge=lfs -text
 samples/v4_768x448_turbo_8steps.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/v3_768x448_16steps_8bit.mp4 filter=lfs diff=lfs merge=lfs -text
+images/comparative.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -102,6 +102,23 @@ Key findings:
   third-party serve-pack conversion (inert `condition_proj`, exploding text-encoder activations),
   not to the pipeline; see `PATCHES.md` and the report in `BIG_GOAL.md` of the source workspace.
 
+## Example outputs
+
+![MiniMax-H3 MLX comparison — 8-bit vs turbo variants](images/comparative.png)
+
+Contact sheet above, prompt *"a red fox leaps over a mossy log in a misty forest"*, seed 42.
+Columns are time (0.4s → 3.0s); rows are the three validated builds: 8-bit 16-step, 8-bit +
+Turbo LoRA, and 4-bit + Turbo LoRA at native 1344x768.
+
+Sample clip (played inline by the Hub) — the 8-bit build this repository ships:
+
+| File | Build | Settings |
+|---|---|---|
+| `examples/v3_768x448_16steps_8bit.mp4` | 8-bit (this repo) | 768x448, 16 steps, 33.9 min |
+
+The turbo variants are not published as video yet (a 1344x768 4-bit + Turbo LoRA re-render with a
+fixed mux is planned); their frames are rows 2-3 of the contact sheet.
+
 ## Turbo LoRA
 
 `turbo_lora_4step_ema.safetensors` is the 4-step EMA distilled LoRA from
@@ -128,15 +145,3 @@ remaining pipeline is byte-identical to the base commit.
   mlx-serve pack note). By downloading you accept those terms.
 - **Port code**: Apache-2.0 (PipeNetwork/minimax-h3-mlx).
 - **Turbo LoRA**: Apache-2.0 (larryvrh/MiniMax-H3-Turbo-Lora).
-
-## Samples
-
-Both clips: prompt *"a red fox leaps over a mossy log in a misty forest"*, seed 42, M4 Max 68.7 GB.
-
-| File | Build | Settings | Visual verdict |
-|---|---|---|---|
-| [v3_768x448_16steps_no-turbo.mp4](samples/v3_768x448_16steps_no-turbo.mp4) | 8-bit (this repo) | 768x448, 16 steps (15 forwards) | 9/10 — coherent fox walking a mossy log, stable background |
-| [v4_768x448_turbo_8steps.mp4](samples/v4_768x448_turbo_8steps.mp4) | 8-bit + Turbo LoRA (merged) | 768x448, 8 steps (7 forwards) | "impressive for a 4-step turbo" — high confidence |
-
-The 1344x768 4-bit turbo render (95% visual on frames) is not included: its local mp4 mux failed
-(script bug, not the model) and the re-render was paused; frames are archived locally.

examples/v3_768x448_16steps_8bit.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52868a877d04b6ea48db9edd8acce36b759143be8def6c053053637e60e3fe6e
+size 1336657