凤希AI - MiniMax视频保存V2

FxAiMiniMaxVideoSaveV2 (凤希AI - MiniMax视频保存V2) is the stripped-down saver in the fxai-toolkit MiniMax family. Where the v1 saver gives you control over frame counts, FPS, output directory, and audio denoising, V2 hard-codes the common case: all frames go into the video, 24fps, fixed output folder, one transition frame out. Three inputs, three outputs, zero decisions. If v1 is for when you're tuning a segment chain, V2 is for when the pipeline is settled and you just want it to save.

The contract is dead simple. Feed it 图片序列 (decoded H3 frames) and 音频, pick a 视频序号 (-1 auto-numbers the file), and it writes the whole image sequence - every frame, no truncation - into an mp4 with the audio muxed at 24fps, into the fixed sucai directory under your videos folder. The one output you might care about beyond the file: 过渡帧, which is always just the very last frame of the sequence, so the next loop segment can start from where this one ended. That single-frame handoff is the whole long-video continuity story, kept to its minimum viable form.

Outputs: 过渡帧 (IMAGE), 视频文件路径 (STRING), 保存目录 (STRING). Like its v1 sibling, it frees the image tensors, runs garbage collection, and empties the CUDA cache after writing - the pack's low-VRAM hygiene so a long segment chain doesn't accumulate VRAM each pass - and it registers the saved file with the pack's task store for its manager UI.

Honest positioning: this is the "I don't want to think about it" saver. The trade is that everything is fixed - you can't keep the last 22 frames as a transition list (v1's 过渡帧列表), you can't change FPS, you can't denoise the output audio, and you can't pick a custom directory. For a one-shot clip it's perfect. For a careful multi-segment render where you want the tail handoff to be rich (22 frames, matching tail audio) and the naming/location controlled, use v1 - its 过渡帧列表 output is what the v1 encoder's 过渡帧列表 input expects for seamless chaining. V2's single transition frame still works as a first-frame anchor for the next segment, just thinner.

Install: ComfyUI Manager → search "fxai-toolkit", or cd ComfyUI/custom_nodes && git clone https://github.com/fxai666/fxai-toolkit, restart. Auto-installs soundfile/psutil; FFmpeg needed for the muxing; MiniMax family needs the H3 weights and a ComfyUI core with native MiniMaxH3 support. Labels are Chinese (凤希AI; help in QQ group 775649071 and Bilibili; no English Reddit community around the pack itself).

The trap to avoid: because V2 saves every frame and v1 truncates to 生成帧数, a graph that expects the saver to trim to the legal H3 grid will silently produce a video longer than the model's frame grid - which is usually fine for playback but can desync from your audio segment math. Keep the frame count aligned upstream (FxAiMiniMaxFrameCalculate exists for exactly this) and V2 will behave.