How I Got 11M Views and 30K Subscribers in 3 Days With This AI Workflow

What this lesson covers

Karima breaks down her 4-step workflow for creating viral AI portal videos using physical phone footage and reference imagery. She demonstrates how to feed reference images into Claude to structure scene-by-scene prompts before rendering video-to-video transformations in Higgsfield AI with Seedance 2.0.

Key takeaways from the creator

AI-extracted notes, not independently verified product claims. Timestamp links let you check each point in the original video.

1. 00:39 ↗

   Recording real physical base footage with sufficient lighting ensures reliable character visibility and realistic motion tracking in video-to-video pipelines.

2. 01:40 ↗

   Using an LLM like Claude by uploading a reference image and describing the shot transition in plain language generates detailed, scene-by-scene prompts with negative prompts automatically.

3. 04:47 ↗

   Selecting 4K output resolution and toggling native sound effects in Higgsfield AI enhances final render quality.

4. 04:52 ↗

   Seedance 2.0 excels at maintaining character emotion, parsing raw footage directly, and adhering to specific camera movement constraints.

5. 05:32 ↗

   Avoid batching multiple generations upfront; inspect the first single render for camera drift or motion flaws, refine prompt instructions in Claude, and batch only after dialing in the result.

Workflow outlined in the video

1. Record clean base footage on a smartphone (e.g., walking through a doorway) with ample lighting across the subject's face and body.
2. Gather aesthetic environment reference images from platforms like Pinterest to define the target world.
3. Upload the reference photo to Claude and describe the desired transformation chronologically to receive structured positive and negative prompts.
4. Open Higgsfield AI, upload both the raw footage and reference media, and select the Seedance 2.0 model.
5. Paste the generated prompts, set the desired aspect ratio, enable sound effects, and render an initial single test generation.
6. Iterate the prompt in Claude if camera moves (e.g., unwanted zoom) or physical elements (e.g., hair blowing in breeze) need correction, then batch final generations.

Before you use this workflow

These notes describe the source video at its publication date. Model access, pricing, connectors and interfaces may have changed. Check the original source and the provider’s current documentation before installing an add-on, connecting an account or spending credits.

ReelStack has not independently tested this workflow. Preview one representative shot and check motion, continuity and output quality before applying it to a full production. No result, cost saving or model capability is guaranteed.

Explore the tools

How to interpret the numbers

243,183 views captured 27 September 2026. ReelStack recommendation score: 59/100. Formula: radar-v5. This is ReelStack’s calculation, not a YouTube rating or a measure of factual accuracy.

Read our methodology and limitations →