ComfyUI Tutorial for Beginners: Image-to-Video, Motion Control & Upscaling (2026)

Module 5 of Comfy 101 covers video and the core idea that makes it manageable: video is just a batch of stills over time. Comfy doesn't treat a clip as a black box. It decomposes it into frames, lets you edit any one of them with the tools you already know, and stitches it back together. Everything from the first four modules comes to a head here. What's covered: → Video Helper Suite — Load Video, Video Info, and Get Image from Batch to pull a clip apart, read its native frame rate and duration, and extract a single frame → Frame editing — sending an extracted frame to Nano Banana, changing it, and feeding it back into the pipeline → Anchoring the shot — building a hero still, then adjusting camera angle and framing with Qwen Image Edit before any motion is generated → One prompt, six takes — routing a single action through Gemini with custom system instructions, splitting the output with a RegEx Extract node, and generating across Veo 3.1, Kling 3.0, Seedance 2.0, Grok Imagine 1.5, LTX 2.3, and Wan 2.2 → Reading the takes — spotting identity drift, camera problems, and temporal coherence, then picking a winner and taking the prompt off autopilot → Finishing — Create Video, then upscaling with Topaz Astra and Starlight Precise or the open-source SeedVR2 Click the links below to run the workflows: Framing Adjustments: [link] Video Upscaling: [link] Chapters: 00:00 Video is just images over time 00:57 Framing the shot like a director 02:38 The Video Helper Suite nodes 03:41 Loading a video and stepping through frames 05:49 Video Info: frame rate, resolution, and duration 07:27 Pulling one frame with Get Image from Batch 08:55 Editing an extracted frame in Nano Banana 10:39 Rolling takes and watching for drift 11:28 The Framing Adjustments workflow on the Comfy Hub 13:48 Building the hero still with custom system instructions 16:27 Generating and locking the base image 17:17 Camera control with Qwen Image Edit 20:31 The video model lineup 21:33 Writing system instructions per model 22:25 Splitting prompts with the RegEx Extract node 23:13 Running all six models in one pass 24:49 Reviewing takes: Veo 3.1, Grok Imagine, Kling 3.0 26:31 Seedance 2.0 and picking the winner 27:21 Open-source takes: LTX 2.3 and Wan 2.2 29:07 Why upscaling is the last step 30:28 Create Video: audio, FPS, and wiring it up 31:24 Topaz Astra vs Starlight Precise 33:08 Open-source upscaling with SeedVR2 36:26 Matching clip length to source frame count 37:18 Exercise: your own shot, six takes 38:15 Troubleshooting flicker, drift, and failed exports