Docs · Content

Image and video generation and prompt history

How to get an image from a description, repeat a shot you liked, and see the exact prompt behind it.

Updated: November 2, 2026

The Image generation tab turns a description into a picture: write what you want to see, and a few seconds later you get a finished shot. Good for carousel covers, backgrounds and post illustrations.

Generating an image#

  1. Open Image generation in the left menu.
  2. Describe the picture in your own words, the way you would brief a designer: "cozy home office with a wooden desk, warm morning light from the window".
  3. Under the input field, pick your settings (all optional, sensible defaults are already set): - model — GPT Image 2 (the default) or one of four others, see below; - format — portrait, square or landscape; - quality — Standard, Pro or Max; each model has its own levels and prices. Higher costs more credits and gives a more detailed or larger shot; - style — a ready preset with a live preview: photo, cinematic, illustration and others; - count — from 1 to 4 images at once.
  4. Press Create. The button always shows how many credits it will take.

You can also attach your own photo (the + button, paste from clipboard, or drag and drop) and ask to rework it — make a studio photo shoot of this person. In that case your text is treated as an edit to your shot, not as a brand new scene.

Not sure where to start? Press 💡 and Craft will suggest a few descriptions to pick from.

History: where your past images live#

Every generation is saved on the same page and syncs across devices — open the app on your phone and you see the same library as on your computer. The slider in the header changes tile size.

⚠️ Without signing in, images are not saved. A guest can generate a few to try it out, but they will not appear in the library — download them right away.

The prompt behind each image#

Click any image to open it large. Underneath you will see a caption with your own request and the settings used: quality, format, style.

The caption has three actions:

  • Copy — put the request text on your clipboard;
  • Repeat — send the request back into the input field together with its settings (quality, format, style). Tweak a couple of words and generate a variation;
  • Download — save the image.

The Full prompt row reveals what was actually sent to the model. A short description gets expanded into a detailed one: lighting, angle, lens, mood are added for you. That expanded text explains why the shot came out the way it did, and it is worth reading whenever the result differs from what you pictured — usually you can spot exactly which detail was filled in for you.

💡 The fastest way to land the shot you want is not rewriting the description from scratch, but taking an image you liked, pressing Repeat and changing a word or two. Settings carry over, so you are comparing variations fairly.

⚠️ Images created before 10 August 2026 have no stored request. We only started saving it on that day, and there is no way to reconstruct it after the fact — those images show a note instead of a caption. Everything generated after that keeps its request.

Which model to pick#

Credits per image by quality level (Standard / Pro / Max):

Model Standard Pro Max When to use it
GPT Image 2 1 4 12 the default; the best at text inside images
Nano Banana Pro — 10 (~4 MP) 20 (~16 MP) photos with your face from your own shot, complex edits, product shots
Nano Banana 2.1 3 (~1 MP) 4 (~4 MP) 9 (~16 MP) the same family, faster and cheaper, fine for most tasks
FLUX.2 Pro 2 (~1 MP) 3 (~2 MP) 5 (~4 MP) photorealism: people, interiors, food. +1 credit for each attached photo
Seedream 5 Pro 5 (~2 MP) — 9 (~4 MP) dense layouts, infographics, lettering
FLUX 3 4 (~1 MP) 10 (~3 MP) — newest FLUX: composition and detail
Ideogram 3 2 — — text, posters, logos
Recraft V3 3 — — illustrations and brand style
Grok Imagine 2 — — photorealism, fast
Z-Image Turbo 1 — — the cheapest, for drafts

On GPT Image 2 the levels change detail; on the other models they change image size. 1–2 MP is plenty for an Instagram post; 4 MP and up is for print and heavy cropping.

⚠️ Models other than GPT Image 2 on Standard need a paid plan or purchased credits. Without them Smart mode is on: GPT Image 2 on Standard, 1 credit per image. The other models appear in the list marked Solo. The last five models in the table don't take reference photos — to edit from your own shot, use GPT Image 2, Nano Banana or FLUX.2 Pro.

💡 Not sure which one — run the same prompt on two models and compare: each reads the same description its own way.

Video#

Switch the prompt box from "Photo" to "Video" and describe the scene to get a clip from text. To animate a picture, attach your photo with the "+" button or click the clip icon on any picture in the feed and describe the motion ("the camera slowly moves in, snow is falling"). Clips and images share one feed, with an "All / Photos / Videos" switch on top.

You choose the model for each clip, and the length — 5–6 or 10 seconds:

Model 5–6 s 10 s When to use
MiniMax H3 Turbo 8 14 fast and affordable, for most tasks
Grok Imagine Video 23 39 lively scenes
Wan 3.0 20 39 smooth motion, 1080p
Kling 2.5 Turbo Pro 27 55 cinematic
Kling 3 Pro 55 110 sharpest — for the cover and the key shot
Kling 4.0 61 122 newest

A clip takes 1–3 minutes and you can close the page — it shows up in the feed on its own. Video needs a paid plan or purchased credits; if a clip doesn't work out, the credits come back. Clips have no sound — music is easy to add when you publish.

💡 An animated picture can go into a carousel slide — that's how video carousels are made, see the article on carousel styles.

What it costs#

Generation spends credits — your balance is shown in the page header. The price depends on the model and quality — see the table above. The exact price is always printed on the Create button and in the model list, so there are no surprise charges.

Credits are granted monthly with your plan, and you can top up with a pack if you run out. If a generation fails (for example, the model does not respond), the credit is returned automatically.

Not available yet#

  • Editing a finished image with a brush or by selecting an area. Edits can only be requested in words, with the image attached as a source.
Didn't find an answer?Message us on Telegram or email support@craftopen.space — we'll reply and extend the article.
Message on Telegram