A store page and a trailer are two files. A face that appears to say a line the person never said is a third job. Putting all three under one “generator” is how a studio ends up with a capsule that is not the build, a trailer that is not gameplay, and a mouth that is not the team’s.
For your own game, keep the files apart. The still is an image call. The trailer is a video workspace. The borrowed face does not belong in either one.
Which file you are making
| Job | File you get | What you hand over |
| Store capsule or patch header | One still | A prompt, or one image you may use |
| Trailer of your own game | A separate video | An idea, a script, a reference video, or a product URL |
| A mouth fitted to a new line | Not the job below | A face and a voice this article does not set up |
The split is the decision. A longer feature list does not turn the first row into the third.
The still, on the image API
The GPT Image 2.5 Flare API is Topview’s route to OpenAI’s GPT Image 2.5 Flare. The route on the page is openai/gpt-image-2.5-flare. Two modes are listed: text to image, and image editing. Image editing takes one input image plus a prompt. The prompt can run to 4,000 characters. There is no field for an audio track, and the page does not accept a video.
Quality is low, medium, high, xhigh, or max. Medium is the default. Resolution is 1K, 2K, or 4K. 1K is the default. The frame is 1:1, 16:9, or 9:16. A wide capsule is closer to 16:9. A launcher icon is closer to 1:1. The page does not list other ratios.
The call is asynchronous. You submit it, keep the task id, and poll until it completes or fails. Result URLs can expire, so download the file. The API key stays on the server. The Playground is where you look at settings. The control that offers a free try copies the prompt into that Playground, where you confirm the settings. API use is paid from the Platform USD balance. Promotional credits vary. The page does not describe a daily free allowance.
On the text-to-image table, medium quality at 1K is 0.0120 US dollars per image on Topview and 0.0240 US dollars official. The page calls that 50 percent lower. Image editing costs slightly more than text to image at the same setting. The rest of the grid is on the page, including other platforms whose quotes may not match quality or dimensions. Read the row you will send. GPT Image 2.5 Sunburst shares the modes, the parameters, and the current rates on this platform. Matching settings do not decide which picture is better. This section follows Flare.
A store still you have the right to use can be the one editing input. Ask for a change you can check, such as the backdrop or the crop. The output is a new picture. Small type on a button or a patch number can come back wrong. If the words have to match the build players will install, use the real interface for that job. Samples on the page include a film poster, a card-game screen, a pixel sheet, and a manga page. Those are the page’s examples, not art from a shipped game, and not renders made for this article.
Do not upload a screenshot of someone else’s game, a character you do not own, or a player’s face. The API will not turn that still into someone speaking the patch notes. It does not take the footage.
The trailer, on the video workspace
Topview is the workspace for a video after the still is settled. You describe an idea or add a script, a reference video, or a product URL. It plans scenes, selects models, generates assets, and keeps the shots editable on Canvas. An image made earlier can feed a video. A product URL can start a marketing video. It does not publish your store page.
The workspace page also describes presenters and voiceover. That is a layer on a video you are making. It is not a method, in this article, for laying a new sentence onto a player, a voice actor, or a character from a game you do not own. Bring the capsule in only after a person has read the words on it. The trailer is a new file. It is not a capture of the match, and a wrong button label does not become true because the shot moves.
The face this article does not set up
Matching a mouth to audio is a different product from the two links above. Those two pages, as written here, take a prompt or a still, or they take an idea for a video. They are not instructions for making a real person appear to say a line they did not say.
If the sentence is a line your team wrote, and the face is a presenter you have the right to use, that is a decision outside this piece. If the face is a streamer, a hired actor who did not record the line, or a character from a game you do not own, leave it out. A store page does not need that mouth to explain a patch. A trailer of your own game does not need it either.
What you can decide
Use the image API when the store or the patch needs one still, and you can name the ratio and the quality before you spend. Medium at 1K is the default text-to-image row, at 0.0120 US dollars per image. Use the video workspace when the job is a trailer of your own game. Leave the talking face out of both until the face, the voice, and the sentence are yours. The price on the row does not buy a mouth, and it does not make the picture a screenshot of the build.
