Image Generation (Beta)
Generate and edit real images inside your chat — text-to-image and image-to-image with per-plan monthly quotas and built-in content moderation.
Tyrex can create real images inside your chat — not placeholders, not links to a third-party site. Describe a thumbnail, a product mockup, or a poster and your AI generates it, saves it as an artifact you can view and download, and keeps it in the conversation so you can refine it until it's right.
Image generation is currently in beta, part of the Tyrex 1.5 lineup, running on Tyrex's own GPU infrastructure.
During beta, generation is limited to a small monthly quota per plan (see Quotas), and generation is one-shot — your AI doesn't automatically re-render based on what it "sees." To refine an image, you ask for a change and Tyrex generates a new version from your edit prompt.
How to turn it on
Image generation is enabled per conversation from the chat composer:
- Open any chat in the dashboard.
- Tap the Create Image pill above the input box (next to the other composer tools) so it's active.
- Type what you want to create.
The tool is only loaded when the pill is on — if it's off, your AI won't call image generation, so there's no accidental quota usage. The same goes for editing: you can edit images only in conversations where the pill is active.
Text → Image
Just describe what you want. A detailed prompt gets a much better result — Tyrex's image model follows the prompt literally:
"A modern flat-lay product photo of coffee beans and a white ceramic mug on a dark wood table, warm morning light, soft shadows, high detail."
Things worth specifying: subject, setting, style, colors, lighting, composition, and any text you want rendered in the image.
Aspect ratio
You can request an aspect ratio in your prompt. Supported ratios:
| Ratio | Use case | Resolution |
|---|---|---|
1:1 | Social posts, avatars | 1024 × 1024 |
16:9 | Thumbnails, banners, slides | 1344 × 768 |
9:16 | Stories, Reels covers, posters | 768 × 1344 |
4:3 | Presentations, documents | 1152 × 864 |
3:4 | Portraits, product shots | 864 × 1152 |
If you don't mention a ratio, you get 1:1.
Image → Image (editing)
You can modify an image that already exists in the conversation — either one your AI generated earlier or one you uploaded.
Two ways to reference the source image, both described in plain language:
- Edit a generated image — say "change the background of the image you just made to a night sky" and Tyrex edits its own previous output.
- Edit an uploaded image — attach a photo to your message and ask for a change, e.g. "make this product photo look like a neon cyberpunk scene."
You can attach up to 6 files per message in the chat composer. If you attach several images, your AI can see them all — for example, to compare four product photos or edit one of them. When you ask for an edit, be specific about which image you mean ("the second photo", "the one with the red background").
How editing works
The AI takes your source image plus a prompt describing the change — style, colors, elements to add or remove, mood. Tyrex's image model applies changes strongly (roughly 90% strength), so edits should be meaningful and noticeable: it can't do tiny surgical tweaks like swapping one word on a label. For precise adjustments, describe a bigger change or regenerate.
The result
Every generated image is saved as an artifact in the conversation:
- View it inline — the image renders directly in the chat message.
- Open it larger — click the image to preview it in a modal.
- Download it — each artifact has a download link; images are served from Tyrex's CDN, so they load fast and stay available.
The image stays in the chat history, so you can come back, point at it, and ask for the next revision — that's the "generate → look → refine" loop.
Quotas
Generation counts against a monthly quota per account, resetting each month:
| Plan | Images per month |
|---|---|
| Free | 3 |
| Pro | 6 |
| Elite | 9 |
One image = one generation or one edit (an edit of an existing image also consumes quota). When your quota runs out, the AI will let you know and won't attempt further generations that month. Upgrading your plan raises the quota immediately.
Both text-to-image and image-to-image calls consume quota — one call, one image. Failed or blocked generations don't count against you.
Content moderation
Tyrex refuses requests for content that is:
- NSFW or adult — explicit, sexual, or suggestive imagery
- Hateful or violent
- Political figures
- Real identifiable people — no deepfakes or likenesses of real individuals
- Brand logos or copyright-protected content
If your request falls into one of these categories, the AI declines politely instead of generating. This applies to both creating new images and editing existing ones.
Limitations (beta)
- One-shot generation — no automatic vision-based re-render loop yet. Revisions are done by asking for a change; the auto-refine loop is planned for a future phase.
- No upscaling or analysis — Tyrex won't upscale an image or "describe what's in this photo" via the image tool; those are different capabilities.
- Cold start — the image model scales to zero when idle. The first generation after a pause can take a minute or two while the GPU container boots; subsequent generations are fast.
- Strong edits only — image-to-image can't do surgical micro-edits (see Image → Image).
FAQ
Why is my image taking a while? The image model runs on a GPU that sleeps when idle. The first request after a quiet period triggers a boot that can take a couple of minutes — later requests in the same session are quick.
Does editing use my quota? Yes — every image-to-image edit counts as one generation against your monthly quota.
Can I use a generated image commercially? Generated images are yours to use, subject to Tyrex's Terms of Service and the content policy above.
Which plan do I need for more images? Quotas are 3 / 6 / 9 per month on Free / Pro / Elite. See Plans & billing for upgrading.