IAMCCS-SuperNodes AU+IMG2VID Exec Render

This is the node that makes the SuperNodes pitch real. A full LTX 2.3 audio+image-to-video run - conditioning, audio VAE, latent planning, segment stitching, continuity, second stage - normally spreads across half your canvas. AU+IMG2VID Exec Render compresses that into one node with a control surface, and it's where the actual sampling happens in the three-stage SuperNode chain (Render → VAE → Finalize).

It's a wrapper, not a replacement: underneath it orchestrates ComfyUI's native LTXV nodes - EmptyLTXVLatentVideo, LTXVConditioning, LTXVImgToVideoInplace, LTXVAudioVAEEncode, BasicScheduler, SamplerCustomAdvanced, and friends. The author's framing (and the community thread that announced SuperNodes) is honest about this: "nothing magical, just a way to make the system usable if you care about structure." If one underlying node is missing or outdated, the Render stage will fail at validation even though it loads fine.

The inputs that matter (and there are a lot)

The full schema is enormous, so here's the tier that beginners actually set:

• generation_type - audio+image2video (default), text+audio2video, img2video, text2video. This picks the pipeline shape.
• ui_preset - the cheat sheet. Presets like low_ram_safe, balanced, high_quality, fast_preview fill a bunch of the scary knobs for you. Start here.
• backend_mode - auto (default) or explicit like two_segments_normal_vram, loop_low_ram_disk. auto is the right first choice.
• positive_text / negative_text - your prompt.
• width / height / steps / cfg / sampler_name / seed - standard sampler controls (defaults: 1280×720, 8 steps, cfg 1, lcm sampler).
• max_shift / base_shift / sigma_terminal - LTX-specific scheduling params.
• image_strength (default 0.8) - how hard the source image anchors the video. Too low and identity drifts; too high and nothing moves.
• audio_context_mode - left_context_only (default) means audio context comes from before the segment, the natural choice for sequential generation.
• vae_mode - inherit_render_backend (default) or explicit tiled/low-RAM modes.
• motion_intensity (default 1) - motion amplification; the community thread's author recommends lowering CFG/image strength to tame exaggeration rather than cranking this blindly.

Outputs

• frames_dir - where rendered frames land (feed to the VAE or Finalize stage).
• segments_rendered, estimated_duration_seconds - sanity numbers.
• linx and report - metadata and status.
• taeltx_preview (optional) - a low-cost preview image when taeltx_preview is on.

Install & first run

Install via ComfyUI Manager (search "IAMCCS") or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart. Requirements: a recent ComfyUI with native LTXV audio-video nodes, your LTX 2.x audio-video model + Video VAE + Audio VAE, Python ≥ 3.12, Torch ≥ 2.8. MelBand RoFormer nodes are optional but recommended for vocal-focused lipsync (if you skip them, use the planner's raw_audio_only mode).

Where people get burned

The requirements doc lists the failure modes: missing deps validate as missing nodes or skipped backends; duplicated pack folders break startup; old saved workflows show shuffled dropdowns after updates. The big one for quality is drift - frames distorting as generation continues. Usual culprits: segments too long, image_strength too low, motion_intensity too high, or anchor refresh off in multi-segment routes. For a first run, stick to the reference recipe: generation_type simple, ui_preset custom, backend_mode single_best or auto, vae_mode normal_tiled_vhs, second_stage_mode off. Confirm that works before touching segments.