2 Commits

Include renames

Author	SHA1	Message	Date
Will Lin	5f1d2ef7d2	[feat] v2: port Wan2.1-T2V-14B (bucket B) + document the all-models backlog ... First bucket-B port toward 'support every fastvideo model': Wan2.1-T2V-14B reuses the Wan recipe + torch adapter unchanged (same WanTransformer3DModel/AutoencoderKLWan/UMT5) — only a registry entry + build_wan_t2v_14b_card (720p, flow_shift 5.0) + SamplingDefaults differ. Without the entry the arch fallback would give it the 1.3B 480p defaults; the explicit entry gives 50 steps / 720x1280. Also recorded the full backlog in V2_PORTING_STATUS.md: 63 fastvideo models = 8 ported / 21 bucket-B (reuse Wan/Causal/LTX2 arch — registry+recipe+defaults, no new adapter) / 34 bucket-C (13 new architectures needing a TorchComponent adapter). Updated the stale 'how to add a model' steps to the post-redesign structure (v2/recipes/, torch_backend.py, SamplingDefaults). CPU mini 211 pass, 2 skip.	2026-07-04 17:15:34 +00:00
Will Lin	1cb2b4e84c	[feat] v2: per-model sampling defaults on ModelCard (Phase 1a) ... v2 had no per-model defaults — generate_video hardcoded 30 steps/25 frames/480x832/cfg5/16fps for every model, badly wrong for e.g. LTX-2 distilled (wants 8 steps @1024x1536) or Wan2.2-TI2V (704x1280@24fps). - New SamplingDefaults dataclass + ModelCard.sampling_defaults (v2/card/specs.py), exported from v2.card. - Populated all 7 supported cards from fastvideo's InferencePreset defaults (steps/guidance/HxW/frames/fps + per-modality guidance for LTX-2.3 A/V). Negative prompts copied verbatim into v2/recipes/_prompts.py (recipe DATA, not model code -> v2-owned, no fastvideo import). - VideoGenerator stores the resolved card; generate_video applies card defaults with precedence kwargs > SamplingParam > card > generic fallback (pure _resolve_default helper, unit-tested). - test_sampling_defaults.py: per-card values + precedence (incl. empty-neg-prompt edge). CPU mini 210 pass, 2 skip.	2026-07-04 17:15:34 +00:00