Day 04 · May 3, 2026

A quick flick transition

Comfy Cloud · Seedance 2.0 · first/last frame

"Interesting transitions don't require fancy gear. They require good shot planning."

Day 4 is a short Sunday family-day episode: record the flick yourself, give the AI model a strong first frame and a clean landing frame, then stitch the generated bridge back into the edit.

You don't need to read this

If you have Claude Code, install the Day 4 plugin and ask Claude to build a flick transition between two iPhone clips.

$ /plugin marketplace add andrewjiang/andrewislearning

$ /plugin install day-04-flick-transition@andrewislearning

What the plugin does

The plugin is not just a prompt. It turns the transition into concrete shot planning and file prep.

Plan gesture

Choose the flick, swipe, or lens-cover frame that can hide the cut.

Pick landing

Use the first clean frame of the next clip as the model's final frame.

Extract frames

Exports exact stills from the same color path as the edit.

Generate bridge

Uses a first-last-frame video workflow in Comfy Cloud.

Stitch

Trims the generated shot short and places it between the real clips.

Color check

Catches the iPhone HDR/HLG yellow-green shift before publish.

The guide

Extract the first and last frames, then send both to Comfy Cloud's Seedance 2.0 first-last-frame workflow.

$ ffmpeg -y -ss 00:00:01.00 -i outgoing.mov -frames:v 1 first_frame.png

$ ffmpeg -y -ss 00:00:00.00 -i landing.mov -frames:v 1 last_frame.png

Prompt the model for a fast, continuous 9:16 handheld move that starts exactly from the flick frame and lands exactly on the next shot.

Prompt: Vertical 9:16 handheld iPhone-style transition. The camera rapidly pulls through the flick gesture, moves outside the house, shows a quick cinematic exterior/roof/neighborhood beat, then dives back into the exact final frame. Realistic, fast, continuous, no text, no logos, no extra people.

The API path

For automation, Comfy Cloud exposes upload, prompt submission, job status, and file view endpoints. Export the workflow API JSON from Comfy, upload both images, submit the prompt, then poll the job until outputs are ready.

$ curl -X POST "$BASE_URL/api/upload/image" -H "X-API-Key: $COMFY_CLOUD_API_KEY" -F "image=@first_frame.png" -F "type=input"
$ curl -X POST "$BASE_URL/api/prompt" -H "X-API-Key: $COMFY_CLOUD_API_KEY" -H "Content-Type: application/json" -d @workflow_payload.json
$ curl "$BASE_URL/api/job/{prompt_id}/status" -H "X-API-Key: $COMFY_CLOUD_API_KEY"

The source

github repo plugin source full guide Comfy Cloud API docs

Next - Day 05

The pipeline keeps getting more visual, but the rule stays the same: only phone, laptop, and AI agents after recording.