Reference Images: Generate From Your Own Photo, Not Just a Prompt
Av The Fellowi Team · · 5 min läsning

Text-to-image generation starts from nothing but a description. Reference images start from something real: a photo you already have, or a companion's existing shot, that the model edits rather than reinvents. If you've ever generated ten variations trying to nail one specific face or one specific outfit, this is the feature that replaces that with "start from this picture instead."
What actually changes
Add at least one reference image to a generation and the request runs on Fellowi's edit model instead of the standard text-to-image path - your prompt now describes what to change or extend about the reference, not what to invent from a blank canvas. You can supply up to 10 reference images in a single generation, drawn from your own uploads, a companion's photo set, or a mix of both.
What it's actually good for
The obvious use is consistency - keeping the same face, the same outfit, or the same character recognizable across multiple generations instead of the model reinventing them each time from a text description alone. It's also the more direct route when what you want isn't a brand-new scene but a change to one you already have: a different setting, a different style, a small edit, starting from a real photo instead of describing every detail of it from scratch in a prompt.
What it costs
Reference images are a small add-on to the generation's normal price, per image, on top of the usual Standard or High cost - current per-image and per-reference pricing is always accurate on the API Console and the product facts page, which both pull straight from what checkout actually charges rather than a number typed into an article. Failed generations are refunded automatically and in full, the same rule that covers every Fellowi Images generation.
Where your past uploads and results live
The studio's generation history keeps a record of what you've generated recently, so a reference image or a result you used last week doesn't have to be re-found or re-uploaded from scratch - it's right there to reuse. Like every generation, results and any reference images tied to them are kept for a limited retention window before deletion; the current retention period is listed on the product facts page.
Try it
Open Fellowi Images, attach a reference photo instead of starting from a blank prompt, and describe the change rather than the whole scene. If you're curious about the pricing model behind the coins themselves, we wrote about that in what AI image generation actually costs.
Läs vidare

Standard eller High Quality? PNG eller JPEG? En snabbguide
Fyra inställningar, en minut att bestämma. Här är när var och en faktiskt spelar roll.

Så skriver du AI-bildprompter som faktiskt fungerar
Vaga prompter ger vaga bilder. Här är vad du ska namnge i stället, med sex riktiga omskrivningar.

Fem AI-bildprompter att kopiera, och varför var och en fungerar
De flesta guider lär ut i fragment. Den här ger dig fem hela prompter att klistra in som de är, och förklarar i vilken ordning delarna hör hemma.