// free downloads
Every node graph I build on YouTube, free to download here. Watch the breakdown, grab the .json, and drop it straight into ComfyUI.
Convert any SDR video to HDR with a clean, easy LTX 2.3 subgraph. Drop in a standard 8-bit clip and it outputs a true high-dynamic-range version with extended highlight and color range, ready to grade or comp. Built as a single subgraph node so it stays simple to drop into bigger pipelines.
HDR AI video that actually fits a VFX pipeline. Image-to-video on LTX 2.3 with a merged graph that outputs proper high-dynamic-range frames and writes EXR, so highlights and color hold up through grade and comp instead of clipping like standard 8-bit AI output. Built for compositors who need linear, scene-referred plates to finish in Nuke.
Replace random AI motion with deterministic control. Direct precise movement for VFX shots (vehicles, characters, fire) using drawn trajectories or Nuke blocking animation. The full optimized node graph for physics-aware AI rendering.
Full or selective AI background replacement for video using Wan and Uni3C. Built for previs and fast-turn projects rather than pixel-perfect VFX, since the AI regenerates edges.
Reverse-engineers what Midjourney and other top platforms do behind the scenes and brings it into ComfyUI for free. Feed a text description or a reference image; the workflow handles prompt engineering automatically and outputs a detailed, cinematically lit, production-quality image. Advanced workflow, learn the ComfyUI basics first.
AI-assisted sky replacement for video that gets about 80 percent of the work done in seconds. Replace the sky on one extracted frame, use UNI3C to transfer the original camera motion onto it, then generate a new clip with rebalanced color, lighting, and shadows. A strong plate to finish in Nuke or After Effects, not a one-click final.
A production-focused VFX workflow using Wan 2.2 Animate for human-centric work where motion fidelity, facial consistency, and lighting matter. With automated roto, pose tracking, and a face-locking system: replace just the face or head while keeping body motion, do full character swaps, de-age, or upgrade wardrobe and props. An AI-assisted starting point for real pipelines, not a final-pixel solution.
A VFX-style clean plate generator that paints out actors, props, boom mics, and clutter automatically. Built on Wan 2.1 VACE for video consistency, QWEN VL3 for auto prompts, Florence 2 for segmentation, and SAM2 for refined, stable masks. Produces clean, stable backgrounds for DMP, projections, and fast patch fixes.
The best SeedVR2 upscaling settings for any GPU, from 6GB to 32GB+ VRAM. Tested across multiple cards to find the right batch size, tiling, overlap, prepend frames, FP8 vs 7B model, and torch compile per VRAM tier. Cleaner images, less flicker, faster renders, fewer VRAM crashes, and better 1080p to 4K upscales.
Selective skin and detail enhancement using a Flux SRPO inpaint setup. From the upscaler comparison video: adds pores, texture, and realistic edges without plastic skin. Part of the image detailer and skin detailer set.
An upscaling reconstruction workflow using Flux DyPE. From the upscaler comparison video: one of the surviving champions for clean, realistic detail without over-sharpening.
Still-image compositing built on Qwen Image Edit 2509 plus Fusion LoRA, optimized for plate-preserving, detail-safe professional use. Keeps the background fully intact, affects only masked regions, auto-matches lighting and shadows, and preserves detail and color. Built for DMP artists, concept artists, and VFX compositors.