Text-to-video iteration checklist
A disciplined loop for improving AI video generations: prepare, test, diagnose the failure, change one thing, and log what worked.
Fill in your details
Values are inserted into every prompt below. Nothing you type leaves your browser.
2 field(s) left — unfilled variables stay as {{placeholders}}.
1Before you generate
☐ One sentence describing what the shot must communicate. ☐ Aspect ratio and duration set in the tool. ☐ Prompt has: shot size, one camera move, subject, one action, setting, lighting, style. ☐ For products or characters: a reference or start-frame image prepared. ☐ A log ready (spreadsheet with prompt, settings, seed if available, result, notes).
2First pass
☐ Generate 2-4 takes from the same prompt to see the model's range before changing anything. ☐ Score each take: composition, motion quality, subject accuracy, style match (1-5 each). ☐ Name the single biggest problem.
3Diagnose and change one variable
Common failures and the variable to change: subject morphs → shorten duration, simplify action, use image-to-video. Camera ignores instruction → move camera words to the start. Too much motion → specify 'slow', remove secondary actions. Wrong mood → change lighting and time of day first, not adjectives. Generic look → add specific materials and imperfect details. ☐ Change only one variable. ☐ Regenerate 2 takes. ☐ Log the result.
I'm iterating on an AI video generation. Original prompt: {{original_prompt}} What went wrong: {{what_went_wrong}} Diagnose the most likely cause in the prompt. Then give me exactly one change to test next (show the revised prompt with the change marked in [brackets]) and say what result would confirm or rule out your diagnosis. Do not rewrite the whole prompt.4Stop and decide
☐ Stop after about 5 rounds on one shot; if it still fails, redesign the shot (simpler action, different angle) rather than prompting harder. ☐ Save the winning prompt to your prompt library with notes on the tool and settings.
Example input
Prompt: 'A golden retriever runs across a beach chasing a ball, camera tracking, sunset, cinematic.' Problem: legs merge and camera stays static.
Example output
Likely cause: the camera instruction is weak and late, and "runs... chasing a ball" asks for fast, complex leg motion. Change to test: "[Steady tracking shot moving right alongside] a golden retriever runs across a beach chasing a ball, sunset, cinematic." Confirms: if the camera now tracks, the order and phrasing were the issue. Leg merging may remain; test "lopes" or a shorter duration next.
Troubleshooting
- If every change seems to make things worse, go back to the best logged prompt and change a different variable.
Known limitations
- Results are partly random; two takes per change is a minimum, not proof.