Hunyuan-DiT v1.1 (Distilled)
Image Available Full comparison ↗ tencent-hunyuan/hunyuandit-v1.1-diffusers-distilled · by Tencent-Hunyuan
· Diffusion transformer (DiT)
· Built on tencent-hunyuan/hunyuandit-v1.1
Pricing — 1 offering(s)
About this offering: The distilled Diffusers variant of Tencent's Hunyuan-DiT v1.1, hosted on Replicate. Distillation targets roughly 50% faster inference than the base model.
Default
- $0.000975 / second
Current
2026-08-22 → present · Nvidia L40S
Replicate bills per second of Nvidia L40S GPU time ($0.000975/s). Its model page estimates about $0.16 per…
Replicate bills per second of Nvidia L40S GPU time ($0.000975/s). Its model page estimates about $0.16 per run (about 6 runs per $1) for a typical ~3-minute prediction, but run time and cost vary with the input.
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Hunyuan-DiT v1.1 (Distilled) fits into a cost-aware routing setup
See how →Capability profile
How Hunyuan-DiT v1.1 (Distilled) rates across core capability dimensions, with the task-level evidence behind each rating.
Model card's headline strength is fine-grained Chinese text/character rendering, a long-standing weak spot for most Western-trained models.
Bilingual CLIP+T5 encoders give solid instruction following; not independently benchmarked against 2026-era frontier models.
Distilled variant specifically targets ~50% inference acceleration over the base Hunyuan-DiT checkpoint.
2024-era DiT model; competent but below current frontier photorealism leaders.
Operator guidance
Route here specifically for Chinese-language or bilingual prompts and in-image Chinese text, where it has a genuine edge most Western-trained models lack. For general-purpose or English-only generation, newer/faster distilled models (e.g. FLUX.1 [schnell], Z-Image-Turbo) or Tencent's own newer HunyuanImage 3.0 are stronger defaults as of 2026.
Use cases
- Chinese-language or bilingual Chinese/English text-to-image generation, especially where in-image text legibility matters
- Cost-sensitive generation via Replicate's per-second GPU billing where the distilled speedup lowers cost per image
- Multi-turn, dialogue-driven iterative image refinement
Limitations
- Released June 2024 (v1.1) — an older generation model; Tencent has since shipped HunyuanDiT v1.2 and the unrelated, much larger HunyuanImage 3.0 line, neither yet tracked in this registry
- No independently published GenEval/T2I-CompBench/FID score found for this specific distilled variant; capability ratings here are qualitative, drawn from the model card and README rather than benchmark data
- Actual per-run cost varies with input/output size since Replicate bills by GPU-second, not a flat per-image price