FAQ
Short answers to the questions people ask first. The words used here are defined in Concepts.
Why reference images and not a LoRA?
Section titled “Why reference images and not a LoRA?”A LoRA teaches one open model one face, product or style. Scenri is the workspace around a model instead: which products, people and places a brand has, which shots worked, and how to make the next one match. It trains nothing. It sends your saved photos with each shot as reference images, so a new person or product is usable the moment it is saved, with no dataset and no training step, and nothing needs retraining when a better model ships.
A LoRA is the better tool when you want open weights running on your own hardware with no per-image fee, a seed you can pin to reproduce a picture exactly (in Scenri, Try again is always a new take), a locked style, thousands of reuses of one subject, or many angles of it. If you already run trained LoRAs on your own machine, Scenri will not beat that on faces.
Neither approach makes logos and small text exact. Scenri redraws products from their photos too; see Is the product in the picture exact?
Why not just use ChatGPT?
Section titled “Why not just use ChatGPT?”The default engine is Codex CLI, OpenAI’s own tool, drawing on the same ChatGPT plan you already pay for. What Scenri adds is around the picture: a saved library of the brand’s products, people and places, a one-line prompt that attaches the right reference photos every time, and each shot’s recipe and refinement trail, so the tenth shot matches the first. If you make one image a week, ChatGPT on its own is enough.
Which engine should I use?
Section titled “Which engine should I use?”| Engine | Needs | Reference images per shot |
|---|---|---|
| Codex CLI (the default) | a paid ChatGPT plan; Scenri installs Codex CLI and signs you in | up to 5 |
| OpenRouter | an API key; draws with google/gemini-2.5-flash-image | up to 4 |
| Replicate | an API token | none |
| fal | an API key | none |
Use Codex CLI if you have a paid ChatGPT plan, and OpenRouter if you do not. Replicate and fal take no reference images, so Scenri refuses a shot that names a product or a person on them rather than drawing a product or a face that only looks right: the refusal says the engine cannot carry the reference images and asks you to choose another engine or remove that chip. A logo on those engines rides as words only, and the composer says so before you send.
Scenri needs Codex CLI 0.157.1 or newer; troubleshooting covers updating it.
What does it cost?
Section titled “What does it cost?”Scenri charges nothing and sells no credits. You pay the provider directly.
- Codex CLI counts against your ChatGPT plan’s Codex usage. OpenAI meters that, not Scenri, so spend caps do not apply to it, and Scenri records Codex shots at no cost.
- OpenRouter bills you itself. For each image Scenri records the cost OpenRouter reports, and about $0.04 an image when it reports none.
- Replicate is about $0.003 an image and $0.04 an edit.
- fal bills per megapixel: about $0.006 an image at Scenri’s sizes, and $0.025 an edit. Scenri’s cost ledger currently records fal images at $0.003, so fal’s own bill runs higher than the figure Scenri shows.
Set a monthly cap for each paid engine in Settings, Providers. Generation stops before a cap is crossed.
Is the product in the picture exact?
Section titled “Is the product in the picture exact?”No, and Scenri does not pretend it is. A new shot is generative reproduction: Scenri sends the product’s own photos (and your logo, when its chip is in the prompt) as reference images, and the model redraws them. Scenri guarantees the right pictures reach the model, not that the model’s rendering is pixel-faithful, so a product can come out slightly different from its photos and lettering can change shape. The full statement for logos is in Brand kit.
Two operations do keep original pixels, restored on your computer rather than redrawn by the engine. Refining one part of a finished shot (adding a prop, removing an object) keeps everything outside that change exactly as it was. Changing a shot to a wider or taller shape keeps the original photograph and draws only the new margin, wherever that leaves no visible join; a tighter shape is a crop.
Can I use the pictures commercially?
Section titled “Can I use the pictures commercially?”Two sets of terms apply, and this is not legal advice. What you may do with a picture an engine drew is governed by that provider’s terms: OpenAI for Codex CLI, and OpenRouter, Replicate or fal and the model behind them for the others. Scenri’s bundled library imagery is covered by ASSETS-LICENSE.md: free to use within Scenri, commercial work included, and everything you generate with it is yours; the library itself may not be redistributed.
Does it work offline? Does it send my data anywhere?
Section titled “Does it work offline? Does it send my data anywhere?”Scenri runs on your own computer, and the desktop icon starts it without an internet connection. Making a picture needs the provider, so generation needs the internet.
When you generate, the prompt and its reference images go straight from your computer to the one engine you connected, and nowhere else. On its own behalf Scenri makes, in the words of PRIVACY.md, “exactly two requests, both disclosed in the terminal the first time they run, and both optional”: a version check against npm every six hours (with the download of a newer release when there is one), and the download of the library imagery archive from GitHub. Both carry no identifier, no telemetry and no user data, and both can be turned off. PRIVACY.md also lists the smaller cases, such as a product’s first photo going to Codex once so it can read the product’s size.
Video?
Section titled “Video?”No. Video is a different craft with different tooling, and it is not planned for 0.x; see the
roadmap.