A real person as the subject reference (Pexels, Oliver Dohrn)

README.md

@@ -20,7 +20,7 @@ This Space is the denoising half of the `ref2va` task: the 61.73 GiB `transforme
 autoencoders. The 62.14 GiB Qwen3-VL conditioner runs in
 [`minimax-h3-conditioner`](https://huggingface.co/spaces/diffusers-internal-dev/minimax-h3-conditioner), which this
 Space calls over the gradio API for every request — the same conditioner Space, and the same resident weights, that
-the keyframe half [`minimax-h3-generator`](https://huggingface.co/spaces/diffusers-internal-dev/minimax-h3-generator)
+the keyframe half [`minimax-h3`](https://huggingface.co/spaces/diffusers-internal-dev/minimax-h3)
 uses.
 
 ## Why split
@@ -55,8 +55,10 @@ Rules the model imposes, enforced here before anything is uploaded:
 
 ## Example assets
 
-`examples/subject.png` is generated (FLUX.1-schnell), `examples/motion.mp4` is a synthetic clip from the parity
-fixtures, and `examples/voice.wav` is utterance `1462-170145-0022` of
+`examples/subject.png` is [a studio portrait by Oliver
+Dohrn](https://www.pexels.com/photo/casual-portrait-of-smiling-woman-in-studio-31747504/) cropped to head and
+shoulders — Pexels licence, free to use. `examples/motion.mp4` is a synthetic clip from the parity fixtures, and
+`examples/voice.wav` is utterance `1462-170145-0022` of
 [LibriSpeech](https://www.openslr.org/12) `dev-clean` — CC BY 4.0, read from a public-domain LibriVox recording. It
 is 16 kHz mono on purpose: the audio VAE wants 32 kHz, so the example exercises the `torchaudio` resample the
 `ref2va` path needs.
@@ -139,6 +141,29 @@ no CFG and no negative prompt (it is guidance-distilled, so every step is one fo
 stops at 14 s because it is the *snapped* count that has to hold for the ceiling: 15 s is 360 frames, which rounds
 up to 362, i.e. 15.083 s, and is refused.
 
+## GPU time is reserved per request, not per Space
+
+MiniMax-H3 attends over one packed sequence, so what a step costs is a function of that sequence's length alone —
+and on this half the references dominate it. A single 1344x768 image reference is ~7168 conditioning rows plus the
+vision block it puts in front of the prompt; a 2.5 s video reference is another ~17000. The same 960x544, 124-frame
+request runs 2.4 s/step with no references and 16 s/step with an image and a video.
+
+`get_duration` prices that before the call instead of reserving a flat ceiling for everything. It takes the
+arguments of the `@spaces.GPU` function, so it has the conditioner's own `text_token_tags` (exact) and the reference
+files (measured from metadata, no decode), and evaluates
+
+```
+S = text rows + reference rows + target rows
+seconds = placement + reference encode + steps * (LINEAR * S + QUADRATIC * S**2) * SAFETY + decode + pad
+```
+
+fitted on the `t2va` half and checked against live `ref2va` requests to about 10%. It matters beyond tidiness: the
+pool reserves whatever number it is given, and a flat 900 s is what makes a busy account fail admission with *"You
+have too many ZeroGPU credits allocated to running tasks."* A typical single-image request now reserves ~460 s.
+
+Every request carries the full placement allowance, because nothing on this side knows whether the worker it lands
+on is cold and a cold one pays the lazy 72.16 GiB `PIPE.to("cuda")` inside its first GPU call.
+
 ## Space variables
 
 | Variable | Default | Meaning |
@@ -146,8 +171,9 @@ up to 362, i.e. 15.083 s, and is refused.
 | `H3_CONDITIONER` | `diffusers-internal-dev/minimax-h3-conditioner` | The Space this one asks for embeddings. |
 | `H3_AOTI` | `0` | `1` loads the compiled block package. |
 | `H3_PLACEMENT` | `lazy` | `lazy` moves all 72.16 GiB onto the card on the first GPU call and leaves it there; `offload` hands placement to `ComponentsManager.enable_auto_cpu_offload` instead. |
-| `H3_ATTENTION` | `_native_cudnn` | cuDNN's fused kernel, 10–20% faster than the SDPA default and needs nothing installed. flash-attention 3 is sm90-only and this pool is sm120. |
-| `H3_GPU_DURATION` | `900` | Seconds per request; the pool applies a 1.5 duration factor. |
+| `H3_ATTENTION` | `_native_cudnn` | cuDNN's fused kernel, 10–20% faster than the SDPA default and needs nothing installed. flash-attention 3 is sm90-only and this pool is sm120. The two VAEs are pinned to torch SDPA instead: they are float32, which cuDNN has no kernel for. |
+| `H3_GPU_DURATION_MIN` / `_MAX` | `120` / `1500` | Bounds on what `get_duration` may reserve. |
+| `H3_PLACEMENT_ALLOWANCE` | `90` | Seconds of the reservation set aside for a cold worker's placement. |
 | `H3_GPU_SIZE` | `xlarge` | ZeroGPU allocation size. `large` does not fit. |
 
 ## Required secret