How to generate better images and videos on Flim
Knowing how to write AI prompts is what turns a generic output into the frame you had in mind. With Flim Pro, the latest AI models generate images and videos from a text prompt, a reference or your own assets, then let you edit, remix and refine them to build mood boards and treatments.
This guide breaks down how to communicate with these models on Flim, so you get better results for your projects.
A formula for effective prompts
1. Build your prompt like a scene
With generative AI, vagueness is the enemy of quality. Anchor every prompt in the fundamentals of storytelling: who, what, where and when. To remove the guesswork, structure it like a shot breakdown:
Prompt = Subject + Action + Environment + Style + Lighting + Camera + Details
2. Be specific, not verbose
Detailed prompts outperform vague ones, but quality comes from precision rather than length. Focus on what matters most: appearance, mood, action, location and atmosphere. These elements have a bigger impact than extra descriptive words.
Flim's tip: Every image generated on Flim already carries a cinematic look and a realistic tone. There is no need to overload the prompt to make it look cinematic.
Prompt:
[Subject: A realistic, collectible action figure of a businessman with a grey suit and lanyard] + [Action: Intensely typing on a keyboard at a computer desk] + [Environment: A miniature cubicle farm office set with high white walls and green carpet] + [Style: Toy photography aesthetic] + [Lighting: Cold, even overhead fluorescent panel lights] + [Camera: Side-profile shot] + [Details: Matte plastic textures, a small coffee mug, a stapler, and a clear toy quality to all materials]
Control the movement
For video especially, describe both subject movement and camera movement. Good results come from clear visual actions rather than static descriptions: "She walks toward the window" is a still, "She walks toward the window, slow dolly in, handheld" is a shot.
To control how a clip starts and ends, use Start / End Frame: provide two images and the model generates the transition. Seedance 2.0 and 2.5 also accept audio references.
Start from a reference image
You can edit any image, whether it comes from Flim's library, your own files or a previous generation. Each aspect of the image can be changed:
Setting: location, time period, season.
Lighting: day to night, soft to hard, cold to warm.
Action: what the characters are doing.
Elements: add or remove characters or objects.
Framing: reframe the shot.
Use Flim's presets
Presets are ready-to-use transformations you can apply to any image, generated or original: camera angle, season, time of day, weather, environment. Sunset relights a daytime still, the Zenithal shot camera preset reframes it from above.
Transfer a style using Aesthetics
To apply the same look across a whole project, build an Aesthetic from 4 to 10 coherent images, or use Flim's ready-made ones. The prompt then describes the subject and the Aesthetic carries the style.
Technical settings on Flim
Before you generate, set your technical parameters in the Flim interface:
Aspect ratio: match the format to the platform, 16:9 or 9:16.
Quality and resolution: from 1K to 4K for images, 720p to 4K for video. Scale up for final renders.
Duration: from 4 to 15 seconds, depending on whether you need a quick transition or a longer narrative shot.
Audio: turn native sound generation on or off.
Model: Nano Banana 2 Lite for fast iteration, Nano Banana 2 and Nano Banana Pro for detail and complex compositions, ChatGPT Images 2.5 for precise edits, Veo 3.1, Seedance 2.0 and Seedance 2.5 for video. Each generation consumes Mana depending on the model, and Pro includes 800 Mana per month.
.png)
Note
Images and videos generated on Flim follow the same rights as the rest of the library: internal pitches, mood boards, research and education only. They are not royalty-free and cannot be used for commercial purposes.
Got
Questions?
Write it like a shot breakdown: subject, action, environment, style, lighting, camera and details. Each part answers a question the model would otherwise guess. Leave out the style part when an Aesthetic is applied.
As long as it takes to be precise, and no longer. Specific choices about light, framing and action matter more than length. If a long prompt produces mixed results, split it into successive edits.
Describe the subject's movement and the camera's movement, then set duration, ratio and audio. For a controlled transition, use Start / End Frame with two images.
For a character or an object, reuse the same reference image in every prompt. For a style, build an Aesthetic from 4 to 10 coherent images, or use a predefined or community one, so the look no longer depends on how each prompt is worded.
No. Generations follow the same rules as the library: internal pitches, mood boards, research and education only. They are not royalty-free and cannot be used in commercial campaigns or client deliverables.