Iterative directing and prompt refinement with Claude
Claude
Demo summary
The user demonstrates 'directing' the AI by telling Claude what was wrong with a previous generation (e.g., character running backwards) to get an updated prompt.
Step-by-step
- Identify specific errors in the previous image generation such as movement direction or facial expressions
- Type corrective instructions into Claude including camera movement styles and character behavior
- Specify elements to keep consistent like color or lighting
- Copy the updated prompt generated by Claude
- Paste the prompt into the image generator to produce a new version
Options
- Add creative camera movements like handheld or Dutch angles
- Specify facial details such as a natural smile or alive eyes
Watch out for
- The user must explicitly state what was wrong in previous batches to fix the prompt effectively
Tips
- Direct the AI like a film director by giving specific movement and framing instructions
- Generate just one image first to verify the prompt updates before committing to a larger batch
Highlights
“The camera is alive. The face holds. We can work with this.”
All demos from “I Made a Cinematic Ad Using Claude Fable 5 + Higgsfield (Full Workflow)”
- 3:350:22Create a product sheet with GPT Image 2.0The user demonstrates how to upload a product image to GPT Image 2.0 and generate a multi-angle product sheet (front, back, and top shots) for visual consistency.Gpt-2· AI Product Photography
- 13:160:39Manage project elements in Higgsfield CollabThe user shows how to add character sheets, locations, and props as 'Elements' in the Higgsfield interface so they auto-attach when pasting prompts.Higgsfield· AI Animation Generator
- 17:390:47Iterative directing and prompt refinement with ClaudeCurrentThe user demonstrates 'directing' the AI by telling Claude what was wrong with a previous generation (e.g., character running backwards) to get an updated prompt.Claude· AI Image Generator
- 27:311:08Create a location scheme using GPT Image 2The user demonstrates drawing dots on a location image and using GPT Image 2 to generate a formal production scheme to fix object positioning in video generations.Higgsfield· AI Storyboard Generator
- Watch “I Made a Cinematic Ad Using Claude Fable 5 + Higgsfield (Full Workflow)” →
AI Image Generator
- 17:390:47Iterative directing and prompt refinement with ClaudeCurrentThe user demonstrates 'directing' the AI by telling Claude what was wrong with a previous generation (e.g., character running backwards) to get an updated prompt.Higgsfield AI
4:340:42Generate consistent environment prompts using ClaudeThe user uploads reference images to Claude to extract visual cues and generate a descriptive prompt for a Japanese dojo that matches the existing cinematic style.Youri van Hofwegen