Avoid conversational requests and over-prompting
Gen-4 generates videos in 5 or 10 second durations based on an input image and text prompt. Users frequently struggle to achieve cinematic results because they treat the model like a conversational partner. You must avoid conversational requests, greetings, or explanations. Instead, use direct descriptions of subject motion, scene motion, and camera behavior. If you request a handheld camera shot while simultaneously asking for a slow dolly movement, the model may produce confused results because it attempts to reconcile disparate instructions that conflict with the physics of the scene. Over-prompting also causes issues. Do not stuff 200 words into a single prompt. Keep prompts under 75 words to avoid diluting instructions. Focus on specific camera terms like "pan," "tilt," or "dolly" rather than vague requests. You should also use simple pronouns or general terms like "the subject" to help the model focus on creating smooth motion. Subject motion describes how characters or objects behave, such as gestures or expressions. Scene motion describes how the environment reacts to movement. You already know the basics of prompting, but these specific pitfalls will save your credits. If your image has multiple subjects, use clear positional language. You might say, "the subject on the left walks forward" while "the subject on the right remains still." You can follow a structure like [Subject description + position] + [environment] + [lighting] + [camera movement] + [mood/style]. Vague words like "beautiful" or "cool" add noise that prevents the model from focusing on the actual scene.
Managing visual consistency and motion artifacts
The reference image system anchors visual identity, but users frequently break this stability through poor prompting. A change of even a single word, such as swapping "brown hair" for "dark hair," causes the model to re-roll facial features and clothing. This results in character drift. If you use the Motion Brush, start with a strength between 30% and 40%. If you set the strength too high, it causes warping and ghosting artifacts.
| Model Feature | Specification |
|---|---|
| Clip Durations | 5 or 10 seconds |
| Available Aspect Ratios | 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 |
| Standard Resolution | 1280×720 |
| Output Frame Rate | 24 FPS |
Combining multiple camera movements, like a dolly combined with an orbit, often leads to background warping or distorted vertical geometry. The model interpolates motion in latent space rather than simulating a 3D camera model. This technical reality means complex, multi-directional instructions can produce unintended results. For multiple subjects, use descriptive identifiers like "the woman nods" or "the man waves" to direct specific motion without confusing the model. You can also use the last frame of a previous clip as the input image for the next generation to chain visual continuity. Can Gen-4 handle simultaneous multi-person dialogue without character drift?
Lighting, subject motion, and technical precision
Lighting direction provides the most control over visual tone. Instead of using vague terms like "cinematic lighting," name a specific source like "morning light from the left." Terminology like "professional lighting" provides no direction, and the model defaults to a flat, ambient output. You should also use positive phrasing. Gen-4 interprets what should happen in a video, not what should be avoided. Negative phrasing produces unpredictable or opposite results. If you want to exclude something, use the separate negative prompt field.
Another common mistake involves re-describing elements that already exist in your input image. Detailed reiterations of the subject’s appearance can reduce motion or cause unexpected results. The image provides the visual starting point for composition, color, and style. Focus your text on describing the movement. For instance, if a subject is already in the image, simply state "the subject turns slowly" rather than re-describing their outfit. You can also choose between insinuated motion and described motion. Insinuated motion uses adjectives, such as "the subject runs across the dusty desert." Described motion provides more emphasis, such as "the subject runs across the desert. Dust trails behind them as they move."
Technical precision in camera language helps avoid generic results. The use of terms like "85mm" for a portrait or "24mm wide angle" for an environmental shot tells the model how to handle depth and focus. For example, an 85mm lens can create a shallow depth of field with soft bokeh. If you use "warm" to describe the skin, wood, and walls in one prompt, the entire frame may take on a yellow cast. Gen-4 is the best tool for creating short cinematic clips when you prioritize motion realism. High-quality input images without visual artifacts yield the best results. Use Gen-4 Turbo for testing compositions and switch to Gen-4.5 for final renders.




