Week 1.2: Consistent AI Video Content
Week 1.2 Resources: Consistent AI Video Content
Required Software & Accounts
Before starting, ensure you have the following accounts set up.
- ChatGPT (Plus/Pro): Required for generating consistent character images, location references, and extracting audio.
- fal.ai: The platform used to run the video generation models. You will need to add billing credits to use the models.
- Minimax H3: The primary video generation model used in the lesson.
- Seedance 2.5 (Optional Alternative): An alternative, more expensive model used for comparison.
Class Prompts
You can copy and paste these prompts directly into ChatGPT or fal.ai to replicate the lesson.
1. Character & Location Generation (ChatGPT)
Create the initial character concept:
create a character image of Bubba, the host of Bubba's kitchen. overweight Italian guy, a bit rough around the edges, but makes a damn good meal every timeRefine the style to be more animated:
make him pixar style(If it still looks too realistic, follow up with: “he still looks too real, make him a bit more pixar”)
Generate the character sheet (Crucial for consistent generation):
give me a character sheet, 16:9, front back left right full body, same angles for face close ups. no textGenerate the consistent kitchen background:
now give me an image of his kitchen, 16:9(If the background is too cluttered, follow up with: “right now it’s way to busy. fix”)
2. Video Generation (Minimax H3 Reference-to-Video)
Note: Ensure you have uploaded your character sheet and/or kitchen image to the “Reference Image URLs” section in fal.ai before running these.
Basic cooking actions:
Bubba prepares a steak in his kitchenBubba prepares a chicken in his kitchenBubba prepares a fish in his kitchenAdding dialogue (Letting the model generate the voice):
Bubba is preparing a fish in his kitchen. He laughs and says "let's cook a fish"Bubba is preparing a burrito in his kitchen. He laughs and says "let's cook a burrito"Forcing a specific camera angle/shot type:
single continuous shot. Bubba introduces himself, saying "Hi, I'm Bubba, and I'm gonna cook a great meal for you." he is standing against a plain white background3. Scripting & Audio Management (ChatGPT)
Extracting audio from a generated video (Upload your video file first):
separate out the mp3 for the audioGenerating a step-by-step cooking script:
give me a step by step simple video prompt for cooking a steak. end with the finished product. include only 5 stepsGenerating an alternative sweet recipe:
do another, something sweet this time