ComfyUI Node
BSAI Prism Sampler (Video + Audio)
A ComfyUI node in BSAI/Prism with 24 inputs and 3 outputs.
BSAI Prism Sampler (Video + Audio)
- prism_model
- image
- audio
- images
- audio
- video_path
◄promptA man with short gray hair plays a red electric guitar.►
◄audio_prompt►
◄negative_promptoversaturated colors, overexposed, static, blurry details, subtitles, watermarks, stylized artifacts, artwork, still frame, washed-out gray, worst quality, low quality, JPEG compression artifacts, ugly, mutilated, extra fingers►
◄resolutionLandscape 720p (1280x720)►
◄width1280►
◄height720►
◄num_frames205►
◄fps24.0►
◄num_inference_steps50►
◄cfg_scale5.0►
◄visual_shift7.0►
◄audio_shift7.0►
◄seed42►
◄enable_vae_tilingfalse►
◄tile_sample_min_size256►
◄tile_sample_stride192►
◄remove_video_dit_after_switchfalse►
◄audio_strength0.80►
◄auto_durationfalse►
◄save_videotrue►
◄filename_prefixBSAI_Prism►
CategoryBSAI/Prism
Inputs (24)
| Name | Type | Default | Description |
|---|---|---|---|
| prism_model | BSAI_PRISM_MODEL | — | |
| prompt | STRING | A man with short gray hair plays a red electric guitar. | Video prompt; supports structured tags <music> <sfx> <speech> <text> <lyrics>. |
| audio_prompt | STRING | Audio prompt; leave empty to reuse the video prompt. | |
| negative_prompt | STRING | oversaturated colors, overexposed, static, blurry details, subtitles, watermarks, stylized artifacts, artwork, still frame, washed-out gray, worst quality, low quality, JPEG compression artifacts, ugly, mutilated, extra fingers | — |
| resolution | COMBO | Landscape 720p (1280x720) | Resolution tier; choose Custom to manually adjust width/height below. |
| width | INT | 1280448–4096 | — |
| height | INT | 720256–2048 | — |
| num_frames | INT | 20517–1024 | Number of frames (pixel space). Auto-aligned to (n-1) % 4 == 0. |
| fps | FLOAT | 24.01–60 | — |
| num_inference_steps | INT | 501–200 | — |
| cfg_scale | FLOAT | 5.01–15 | Classifier-free guidance strength; 1.0 = no guidance. |
| visual_shift | FLOAT | 7.01–20 | Video noise-schedule shift. 720p~7.0, 1080p~9~13, 2K~13~17. |
| audio_shift | FLOAT | 7.01–20 | — |
| seed | INT | 420–2147483647 | — |
| enable_vae_tiling | BOOLEAN | false | Tile VAE decoding (2K/1080p OOM prevention). Tier presets override this automatically. |
| tile_sample_min_size | INT | 25664–1024 | — |
| tile_sample_stride | INT | 19232–1024 | — |
| remove_video_dit_after_switch | BOOLEAN | false | Release the high-noise expert video_dit after the boundary switch to save VRAM (irreversible; internal release only). |
| audio_strength | FLOAT | 0.800–1 | Audio injection strength (lip-sync). 1.0 = fully preserve input audio, 0.0 = ignore audio and generate from scratch. Requires a connected audio input. |
| auto_duration | BOOLEAN | false | Auto-compute frame count from the input audio length (requires a connected audio input). |
| save_video | BOOLEAN | true | — |
| filename_prefix | STRING | BSAI_Prism | — |
| imageopt | IMAGE | Reference image (i2v first frame). Leave empty = text-to-video (denoise from a black frame). Center-cropped and resized to the target resolution. | |
| audioopt | AUDIO | Pre-generated audio (speech/music) for lip-sync. Once connected, use audio_strength to control how much is preserved. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| audio | AUDIO | — |
| video_path | STRING | — |