Pricing
StarGen AI is a Telegram bot for AI images, video and voice, priced in Telegram Stars: one credit is one Star. You see the price before every confirmation, and exactly that amount is charged.
12 Stars per week
Spend it on any model, no card. The Stars land once a week and do not stack: whatever is left when the week ends is replaced by the new week's.
1 Star · one fast image
Plans
Tiers by the month
Starter
250 Starsper month
- 250 Stars credits every month
- 2 jobs at once
- Priority queue
- Unlimited fast images
Plus
750 Starsper month
- 750 Stars credits every month
- 4 jobs at once
- Priority queue
- Unlimited fast images up to 50 a day
Ultra
2,500 Starsper month
- 2,500 Stars credits every month
- 8 jobs at once
- Priority queue
- Unlimited fast images up to 100 a day
Top-ups
Packs of Stars
Purchased credits never expire. It is one balance, not a plan.
Catalogue
The models we run
Every model on the shelf today, with what the cheapest run of it costs. You pick the one you want; a subscription never unlocks or locks any of them.
Images
Google's flagship image model. It draws from a description and reworks whatever you send it, and text inside the frame comes out readable, which most models still cannot manage.
OpenAI's image model reads a long prompt closely and keeps things where you asked for them. Good when the frame needs legible text or a scene built exactly to description.
The quick draft model: you type a prompt and it draws it. Take it when you want to run through a dozen ideas rather than polish one.
The bigger Nano Banana: 2K output, clean text inside the frame, and edits that keep the detail. It works from a photo or from a prompt alone.
ByteDance's photo editing: it swaps clothes, backgrounds and style without redrawing the person. It will not start without a photo, text alone is not enough.
The fastest image here and one of the cheapest. Take it when you want to run through a dozen ideas rather than polish one.
Google's editor: attach a photo, say what to change in plain words, and the face and the light survive it. Leave the photo out and it just generates from the prompt.
The junior Seedream: the same grasp of a complicated scene as the Pro, at a fraction of the price. Strong on keeping a frame internally consistent.
xAI's image model in three quality rungs. The bottom one costs almost nothing; the top one reaches for the flagships.
Black Forest Labs' current line, in three rungs. It bills by megapixel, so the frame here is pinned at exactly one.
The editing model people mean when they say editing: it holds a face and a style from frame to frame, which is why it is used on series rather than single images.
OpenAI's new image models for creating and editing: Flare for quick ideas, Sunburst for precise revisions. Choose quality and canvas size before generating.
Alibaba's flagship. One of the strongest in the world on blind comparisons, and inexpensive with it.
Recraft for design work: clean graphics, careful typography, a result you can predict.
Krea in three rungs, from very fast to large. It has an aesthetic of its own, which is what people come to it for.
Edits a photo you send from a written instruction: swap the background, the style, one detail. The cheapest edit on the shelf.
Raises a photo's resolution and cleans up the noise. Tens of times cheaper than the professional upscaler, because the output size is one you pick rather than one your file decides.
The only model here that returns real SVG rather than pixels. Logos and icons that scale to any size afterwards.
OpenAI's previous image engine, in two quality rungs. Cheaper than the current one and still in the world's top ten.
The senior Recraft, for work that goes into production rather than into a chat.
It enlarges an image four times and pulls detail back out instead of stretching pixels. For old photos and small screenshots you need to print or show large.
Video
ByteDance's flagship: it holds a character and a scene from shot to shot instead of rebuilding them every second. It comes in 720p, 1080p, and an edit pass over video you already have.
The cheapest video here. The draft rung costs less than the free weekly allowance, which means your first video need not cost you anything at all.
First in the world on blind text-to-video comparisons, and a third the price of the flagship we used to lead with.
Google's video model with sound, and the sound is real speech rather than dubbed-in effects. The only model here that does it.
The quick branch of Hailuo: it turns a photo into six seconds of movement, and the prompt is optional. It needs an actual photo, text alone will not start it.
Kling 3 Pro writes the sound along with the clip: footsteps, room tone, atmosphere. It starts from a photo and gives the cleanest motion in the Kling line.
xAI's video model: a photo becomes a five-second scene, and a clip you already have becomes a new version of itself. What you attach decides which of the two you get.
The small Seedance: the same feel for movement as the flagship, in a single 720p rung driven by text. For drafts and for testing an idea before you run the bigger model.
Kling 2.5 Turbo turns a photo or a written scene into a clip of five or ten seconds, faster and for less than Kling 3 Pro. 720p or 1080p from a photo, 1080p from text, no sound.
The same Hailuo, but it starts from text: describe the scene and get six seconds. It keeps the physics honest, so people and objects move the way you expect.
Inexpensive video with sound. The middle ground for when Veo is too dear and a draft is not enough.
Classic Kling: it turns a photo into a five-second clip with a smooth camera move. The workhorse for when you want movement without surprises.
One photo plus a script, and the person in the photo says it. Thirty voices to choose from.
Google's video model for motion and sound. Describe a scene or start from a photo, then choose length, resolution, and audio.
Runway Gen-4.5 makes video from a description or a starting image. Choose the frame shape and a length from 2 to 10 seconds.
Luma Ray 3.2
Luma Ray 3.2 creates video from text or animates a photo. Choose from three resolutions, two text-video lengths, and a seamless loop for five-second clips.
Voice-over
ElevenLabs at its most expressive: you hear pauses, intonation and emotion instead of a robot. For voiceover and anything a person will listen to end to end.
OpenAI's text to speech: an even voice and clear diction, nothing to set up. The plain option for long text, where being understood matters more than acting.
Studio-grade narration with control over delivery and emotion. It sounds good in Russian.
The speed model in the ElevenLabs line: it says short lines cleanly with nothing to configure. Take it for dialogue and reactions, not for half an hour of reading aloud.
Voice cloning
It takes a voice off a recording and turns it into one you can speak text with. It needs a clean sample, no music and nobody else in the background.
FAQ
Questions about pricing
What is a Telegram Star?
Stars are the currency Telegram sells for paying bots and Mini Apps. You buy them inside Telegram, and in StarGen one credit is one Star.
Is StarGen free to use?
Every account gets 12 Stars a week with no card. The cheapest image and voice runs start at 1 Star, so the allowance covers several of those. Unused Stars do not carry over to the next week.
Do I need a subscription?
No. Credits buy any model, and a pack is a one-off purchase that never expires. A monthly tier adds credits every month, more jobs at once and a priority queue. It never locks or unlocks a model.
What happens if a generation fails?
When a generation fails on the provider's side, the credits return to your balance automatically. Refunds of Stars paid to Telegram go through /paysupport in the bot.
Can the prices change?
Yes. Star prices track what the models cost us and can move. The price on the generate button is the price charged, and a change never applies to a generation that is already running.
Star prices track model costs and can change. A change never applies to a generation already running.
Bought outside the app, Stars usually cost a little less.