TLDR

• Magic Mirror Fast runs on Nano Banana with a fixed ratio and quick turnarounds

• Multi image inputs across supported models, up to 10 images

• Telegram bot, deep links, device login, and Stars purchases that credit points

• S3 + CDN delivery for heavy images, plus readiness polling and smarter upscaling

• Async creation batches with status polling and clearer mission cards

• New admin tools for payments, bans, purging, and uploaded images

Magic Mirror Fast on Nano Banana

We added Nano Banana and built Magic Mirror Fast on it. The model keeps a fixed aspect ratio and prioritizes speed, so you get results quickly without extra setup. Costs are shown up front, and a small PayPal mini checkout appears in place if you need a few more points.

• Fixed ratio badge and a disabled ratio picker for Nano Banana

• Inline top up and a clean handoff back to your job after purchase

• Clear names in the UI: Magic Mirror Pro for the personalized path, Magic Mirror Fast for the quick path

Use multiple images as input

Models that support it can now take several uploaded images at once. Selection is capped, deduplicated, and adapts if you switch models.

• Multi image inputs for Nano Banana and gpt-image-1, up to 10

• Single image inputs for flux-dev, flux-pro, flux-pro-ultra, photon, and custom ai models

• Selections auto trim or clear when you switch to a model with different limits

The new picker is quick to scan, has remove and clear-all controls, and uses S3 thumbnails that load well on desktop and mobile.

Telegram bot, login, and Stars purchases

Telegram is now fully supported. Link your account, sign in on desktop with a six digit code, and buy points with Stars. Deep links expire for safety and the flow includes docs, idempotency, and optional webhook secrets.

• Link and buy deep links via API with expiry and optional tracking

• Device login flow: create, confirm with code, exchange for a session token

• Stars purchases credit points to the linked user and apply referral bonuses

• Admin tools to generate ban links and manage bot settings

Faster images, cheaper polling

Heavy images now come from S3 + CDN with short-lived secrets for HD, originals, and upscales. Upscaling uses a readiness endpoint and S3 locks so work is not duplicated.

• HD and upscales over CDN using short secrets, not base64 blobs

• Readiness-only polling cuts bandwidth while waiting

• Best-effort S3 locks and cleanup stop duplicate upscales and stale waits

On the client, the gallery mounts media only when visible, retries are light, and profile pages use proper pagination. Long lists stay stable and fast.

Async creation and missions

Creation can run in the background. The new batch status API reports waiting, running, completed, and error states with per job details like waiting for model, finished counts, and progress. While jobs run, the grid shows placeholders that swap out when assets arrive.

• Batch status includes percent complete and types for image and video jobs

• Placeholders appear during propagation, then auto replace

• Mission cards show clear progress bars, point and badge rewards, and a single claim button

• A once per session prompt lets you know when claims are available

Admin and operations

Moderation and finance got a full pass.

• Media purge removes banned content across all storage variants, supports dry run, batching, and retries

• Admin ban endpoint now returns usage stats at the time of ban

• Payments page with server filtering by method, status, user, and date range

• Admin uploads browser with table or grid view, date filters, rows per page, and ban from the thumbnail

Marketing and analytics

Meta Pixel now tracks purchases, registrations, SPA pageviews, and includes a no JS fallback. Legal links for Privacy and Terms are in the footer and model pages. The PayPal mini checkout renders only when visible and cleans up after itself.

Quality and UX fixes

• Profile pages reliably jump to top when switching tabs or users

• Favorites moved to the same paginated, searchable flow as creations

• Upvoting private media shows a clear dialog and hides the button when privacy is known

• Better loading overlays, error handling, and correct image orientation on upload

Why this matters

These changes make Fiddl.Art faster and easier to navigate. S3 delivery and readiness polling lower costs and improve page speed. Multi image inputs with Nano Banana turn complex edits into a few clicks, like an ai art playground. Telegram login makes sign in simple for chat first users. Admin tools reduce time spent on moderation and reconciliation.

If you rely on custom ai models or run an ai headshot generator workflow, you get clearer inputs, up front costs, and predictable results. If you are comparing the best ai art generator tools, this update focuses on the basics that matter: speed, clarity, and control for text to image ai and ai art with custom models.

Notes for developers

• Model enums now include nano-banana across creation endpoints

• Creation payloads accept uploaded image UUIDs and enforce per model caps

• HD, original, and upscales use CDN URLs with short secrets; a download flag is available

• Batch status API exposes progress and per job states, including waitingForModel

• Telegram supports deep links, device login create-confirm-exchange, and Stars purchases with idempotent crediting

What is not changing

All existing routes and models remain. Legacy base64 endpoints still work, but CDN URLs are now preferred. No roadmap items were added beyond what is listed here.

Thank you

Thanks for the feedback that shaped this release. Tell us how Magic Mirror Fast and multi image inputs fit your workflow, and what you want to see next.