Ce que nous avons publié
Chaque version, amélioration et correction, de la plus récente à la plus ancienne.
- v3.40 Jul 10, 2026
Sharper Videos and Optional Sound
Videos made from an image now have a Max Quality option for extra sharpness, plus optional AI-generated sound. Both are optional; the fast default is unchanged.
Lire la suite Réduire
Max Quality Option
A new Max Quality choice renders video with more detail and sharper faces. It takes a little longer, and the fast default stays the same, so you opt in only when you want the extra polish.
Optional Sound
You can now add AI-generated audio to a video. It is optional and off by default, so your videos stay silent unless you turn it on.
- v3.35 Jul 8, 2026
Sharper Video From The Animate Button
Turning an image into a video with the Animate button now renders a much sharper, higher-resolution clip with clearer faces and finer detail. It uses the fast high-quality video engine by default.
Lire la suite Réduire
Clearer, Higher-Resolution Animation
The Animate button now uses the fast high-quality video engine by default, so image-to-video results come back noticeably crisper with more natural detail in faces and hair.
- v3.33 Jul 2, 2026
Smoother, Sharper Image-to-Video
Videos made from an image now play at their natural, lifelike speed instead of running fast, and they render at a higher resolution with clearer faces and eyes. Motion is calmer and more natural, and the result looks noticeably crisper.
Lire la suite Réduire
Natural Playback Speed
Image-to-video clips previously played back too fast. They now run at their true, intended speed, so movement looks natural instead of rushed.
Higher Resolution And Clearer Faces
Videos now render at a higher resolution with an optional face and eye refinement pass, so faces and eyes come out sharper and more lifelike, and overall motion is calmer.
- v3.32 Jun 23, 2026
Text-to-Video Always Produces a Playable Video
Creating a video from a prompt now always builds a still image first and then animates it, the path that already worked reliably. The older direct text-to-video route was removed because it could return a clip that would not play. Standalone video upscaling is also hardened against expired or unreachable source links.
Lire la suite Réduire
Reliable Video From A Prompt
Generating a video from a text prompt now always creates a still image first and then animates it. The older direct text-to-video route is gone: it could produce a clip that would not play and sometimes failed on a missing model file. Every prompt-to-video request now runs the proven image-first path and yields a real, playable clip.
More Dependable Video Upscaling
Standalone video upscaling now sends the source clip to the upscaler in a self-contained form instead of relying on a temporary link that could expire or be unreachable. An expired or stale source link can no longer make the upscale fail.
- v3.31 Jun 23, 2026
Honest Video Lengths + Reliable Video Generation
Studio video now offers only the lengths it actually delivers (2s, 4s, 5s), so you are never charged more for a longer clip that still comes out around 5 seconds. Image-to-video generation also retries once on a slow start and fails or refunds faster when truly stuck.
Lire la suite Réduire
Lengths You Are Actually Charged For
The video length picker now offers 2s, 4s, and 5s only. Longer options used to be available but the model still produced about 5 seconds while charging the higher price. The picker now matches what you actually get, so the length you pay for is the length you receive.
Automatic Retry On Slow Starts
Image-to-video sometimes failed on the very first attempt when the generator was starting up cold. It now retries once automatically before giving up, so a one-off slow start no longer wastes your credits or your time.
Faster Fail And Refund When Truly Stuck
A genuinely stuck video job is now detected and refunded sooner instead of hanging for a long time. If something does go wrong, you get your credits back faster.
- v3.30 Jun 22, 2026
Studio Formats Match What Generates
The Studio image format picker now offers exactly the formats that generate (1:1, 2:3, 3:2, 4:5, 9:16) on desktop and mobile. 4:5 is now available; formats that could not generate were removed.
Lire la suite Réduire
4:5 Portrait Now Available
The 4:5 portrait format is now offered in the Studio image generator. It was supported behind the scenes but never shown, so you could not pick it. It now appears in the format picker on both desktop and mobile.
No More Formats That Error Out
Formats that the generator could not actually produce (and that returned an "unsupported aspect ratio" error if picked) have been removed from the picker. If you had one saved from before, it is automatically switched to the closest supported format by orientation.
- v3.29 Jun 22, 2026
No More False "Failed" on Slow Videos
High-quality image-to-video clips no longer show a false "failed" while they are still generating, and every aspect ratio (including 4:5) keeps its true shape.
Lire la suite Réduire
Slow Clips Stay In Progress
High-quality videos can take longer to finish, and the gallery used to mark them failed too early. Clips now stay marked as generating for the full time they need, and if one was briefly shown as failed but actually finished, it recovers to completed on its own with no reload.
Every Aspect Ratio Supported
All aspect ratios, including 4:5, now generate at their true shape instead of falling back to a square. The clip you ask for is the shape you get.
- v3.28 Jun 22, 2026
Auto-Sharpen No Longer Shrinks Square Clips
Automatic video sharpening now keeps every clip at least its original size, so square videos stay full size, and clips keep their original playback smoothness.
Lire la suite Réduire
Square Clips Stay Full Size
The background sharpening for image-to-video clips now never makes a clip smaller than the original. Square clips that previously came back at a reduced size now stay at full size, and every sharpened clip keeps the same playback speed and smoothness as the version you first saw.
- v3.27 Jun 22, 2026
Image-to-Video Clips Now Auto-Sharpen
Videos you create from an image now upgrade to a sharper, higher-resolution version automatically in the background, with no extra credits.
Lire la suite Réduire
Sharper Videos, Nothing to Click
When you turn an image into a video, the clip appears right away just like before. In the background we now run a high-resolution upscale and quietly swap the clip in your gallery for the sharper version a few minutes later, in the same spot, at no extra cost. If the upscale ever cannot finish, your original clip stays exactly as it was so you never lose a video.
- v3.26 Jun 22, 2026
Sharper Image to Video, and Your Duration Now Sticks
Turning an image into a video now defaults to a sharper, higher resolution clip, and the length you pick actually changes how long the video is.
Lire la suite Réduire
Higher Quality by Default
When you turn an image into a video, the result now comes out at a sharper, higher resolution by default instead of the lower resolution preview quality. You get a noticeably crisper clip without changing any settings.
Your Selected Length Is Respected
The video length you choose now actually drives how long the generated clip is. Pick a longer duration and you get a longer video, instead of every clip coming out the same fixed length.
- v3.25 Jun 09, 2026
Show or Hide Your Password While Typing
Password fields now include an eye button to reveal what you typed and hide it again, so you can catch typos before submitting.
Lire la suite Réduire
Catch Password Typos Before You Submit
Every password field now has an eye button on the right. Tap it to reveal your password as plain text, and tap again to mask it. It works everywhere you enter a password, including sign in, sign up, password reset, setting your first password, and changing your password in settings.
- v3.24 Jun 09, 2026
Bigger Voice Library With Search
Voice selection now opens the full library. Search by name and filter by gender or vibe across the wizard, settings, and chat.
Lire la suite Réduire
The Whole Library, Everywhere You Pick a Voice
Voice pickers used to show a short fixed list. Now you can reach thousands of voices in the character wizard, your influencer settings, and chat. Search by name or filter by gender and vibe to narrow it down fast.
A Focused Start, The Full Set On Demand
The wizard still opens with a focused set of voices matched to your character and the vibe you pick, so you are not scrolling thousands on the first screen. Tap "Browse all voices" or clear the filters any time you want the complete library and search.
Clear Message When A Saved Voice Goes Away
If a voice you saved earlier is no longer available, settings now tells you clearly and prompts you to pick a new one instead of showing a generic error.
- v3.23 Jun 08, 2026
Clearer Face Model Training Start
Starting a face model now confirms it kicked off, tells you it takes about 1 to 2 hours, and emails you when it is ready.
Lire la suite Réduire
You Now Know Your Training Actually Started
When you start training a face model, you now get a clear confirmation that it kicked off, along with the expectation that it takes about 1 to 2 hours. We email you and notify you in the app the moment it is ready, and the settings panel shows a live in-progress state while it runs. If a start cannot go through, for example because you have reached your monthly training limit, you now see the exact reason instead of a generic error.
- v3.20 Jun 08, 2026
Failed Trainings No Longer Count Against Your Monthly Limit
Failed AI face model trainings no longer count against your monthly training limit. If a training fails you got nothing, so it will not use up one of your monthly slots.
Lire la suite Réduire
Failed Trainings Stop Using Up Your Slots
Previously, an AI face model training that failed still counted toward your monthly training limit, so a run of failures could block you with a "you have used your trainings this month" message even though you received nothing. Now only trainings that are in progress or that produced a usable model count toward the limit. Failed trainings are ignored, so you keep your full monthly allowance.
- v3.19 Jun 06, 2026
Avatar Videos Now Finish Reliably
Fixed a bug where avatar videos could get stuck and never finish, leaving you waiting with no result. Avatar videos now complete in about two minutes, and if a generation cannot finish your credits are returned automatically.
Lire la suite Réduire
Avatar Videos Complete in About Two Minutes
Avatar videos previously could sit in processing indefinitely and never deliver a result. The completion handoff has been fixed, so a talking avatar video now finishes in roughly two minutes and appears in your gallery. If a generation ever fails, your credits are refunded automatically and you get a clear message instead of a silent hang.
- v3.18 Jun 06, 2026
Studio Now Starts From Templates
The Studio opens with a template-forward surface. Pick your character, tap a template image, and generate in one step. The prompt is hidden behind Advanced, and on phones each category opens a full-screen picker.
Lire la suite Réduire
Tap a Template, Generate Instantly
The Studio now leads with templates: a hero template plus horizontal category rows of large images sit above the generation bar. Select your character, then tap any template image to apply its full configuration and generate in a single step, no prompt writing required. The freeform prompt is tucked behind an Advanced option for anyone who wants to fine-tune.
Signature Templates and Full-Screen Mobile Picker
Signature templates show a lock badge and keep their prompt private while still generating great results. On phones, browsing a category opens a full-screen picker with large, easy-to-tap cards, and the bottom navigation stays in place.
- v3.17 Jun 05, 2026
AI Face Model Images No Longer Show the Wrong Face
Fixed a bug where studio images for a character with a trained AI face model could come back showing a completely different person, sometimes the wrong gender. Generations now stay true to your character.
Lire la suite Réduire
AI Face Model Images No Longer Show the Wrong Face
Fixed a bug where studio images for a character with a trained AI face model could come back showing a completely different person, sometimes even the wrong gender. The face model was being applied with the wrong rendering engine. Generations now use the engine that matches how the face model was trained, so the face stays true to your character. If a face model ever cannot be matched to a compatible style, the generation now stops with a clear message instead of returning a wrong face.
- v3.16 Jun 05, 2026
Character Sheets Keep Your Character's Face
Character sheets now preserve your character's face and identity on every shot, matching the base image. Regenerated sheets stay consistent too.
Lire la suite Réduire
Character Sheets Keep Your Character's Face
Character sheets now preserve your character's face and identity on every shot, matching the base image you started from. Previously some sheets, especially regenerated ones, could come back without consistent facial features. Every sheet is now identity-consistent.
- v3.15 Jun 05, 2026
Sharper Full-Body Framing + Auto-Retry for Face Model Training
Full-body and wide studio shots now use a portrait canvas so the whole figure fits without cropping or zooming. Face model training that dies mid-run now retries automatically at no credit cost.
Lire la suite Réduire
Full-Body Shots Are No Longer Cropped
When you pick a full-body, cowboy, wide, or similar head-to-toe shot, the studio now generates on a portrait canvas automatically. Previously a square or landscape canvas could crop the figure or zoom in on the face, so the pose you asked for was cut off. Standing poses now show the whole subject in frame.
Face Model Training Auto-Retries
If a face model training run stalls or dies before finishing, the system now detects it and retries automatically, with the retry free of charge. You no longer have to manually restart a stuck training job.
- v3.14 Jun 05, 2026
AI Face Model: Training No Longer Fails Early
Fixed a bug where AI face model training that runs longer than a few minutes could be marked as failed before it finished. Training now has time to complete.
Lire la suite Réduire
AI Face Model: Training No Longer Fails Early
Fixed a bug where AI face model training that runs longer than a few minutes could be marked as failed before it finished. Training now has time to complete, so your face model is created reliably.
- v3.13 Jun 03, 2026
Subscription Cancellation Now Reliable
Fixed a bug where cancelling a subscription did not always take effect, which could lead to an unexpected renewal charge. Cancellations now apply correctly every time.
Lire la suite Réduire
Subscription Cancellation Now Reliable
Fixed a bug where cancelling a subscription did not always take effect, which could lead to an unexpected renewal charge after you had already cancelled. Cancellations now apply correctly every time and you will not be charged again after cancelling.
- v3.12 Jun 3, 2026
AI Face Model Training No Longer Gets Stuck
Training jobs that do not complete within 30 minutes now automatically reset so you can retry. A duplicate-job guard prevents accidentally queuing two runs at once.
Lire la suite Réduire
AI Face Model: Training No Longer Gets Stuck
Fixed a bug where AI face model training jobs could remain stuck in "Training..." indefinitely with no way to retry. Jobs that do not complete within 30 minutes now automatically reset to failed so you can start a new training run. A duplicate-job guard also prevents accidentally queuing two training jobs at the same time.
- v3.11 Jun 1, 2026
New Chat Design + Model Consistency Indicator
Chat messages now appear in clean character and user bubbles with a persistent header. Studio model selection shows which models have strong face consistency for your character.
Lire la suite Réduire
Chat: New Conversation Design
The chat interface has been redesigned. Messages from your character appear in a dedicated bubble with their avatar and name. Your replies are cleanly separated on the right. A persistent character header stays visible as the conversation scrolls. The input area is fixed at the bottom.
Studio: Model Face Consistency Indicator
The model selection modal now shows which models support strong face consistency for your character. Pick the right model for a given shot without guessing.
- v3.10 May 25, 2026
Chat: Conversations Now Work Reliably
Fixed a crash that showed "Something went wrong" when sending chat messages. AI replies are now stable with conversation memory working within each session.
Lire la suite Réduire
Chat Page Crash Fixed
Sending a message no longer causes the chat page to show an error screen. Conversation memory within a session works correctly; the AI recalls what you said earlier in the same conversation.
- v3.9 May 19, 2026
Chat Agent: Create Characters by Conversation
Describe the character you want in chat and the AI creates it. Studio image generation is also available via conversation; a progress indicator shows while generating and the image appears inline.
Lire la suite Réduire
Create Characters by Conversation
Instead of navigating menus and filling forms, describe the character you want. The AI agent creates it and shows a character card with name, archetype, and quick-action buttons.
Trigger Studio Images from Chat
Ask your character to generate a studio shot. A progress indicator shows while the image generates, then the completed image appears inline in the conversation.
- v3.8 May 15, 2026
Faster Studio Generation via Dedicated GPU
Studio generations now route to a dedicated GPU pod for lower latency at peak load. Modal remains as automatic fallback. No user action required.
Lire la suite Réduire
Dedicated GPU Pod
z-image-simple generations now route to a dedicated GPU instead of cold-start serverless, reducing time-to-first-image at peak load. If the pod is unavailable, requests automatically fall back to Modal with no interruption.
- v3.7 May 14, 2026
Studio Gallery, Wizard, Voice and Rewards Fixes
Studio gallery reliably loads images on desktop at any screen width. Rewards tab restored in sidebar. Wizard character face generation works for all 6 slots with automatic retry. Voice selection no longer errors after multiple regenerate clicks.
Lire la suite Réduire
Studio Gallery Restored
Gallery now loads images reliably on desktop at all screen widths. A React hydration issue was causing the image grid to render empty on first load.
Rewards Tab Visible
The Rewards tab is now visible in the sidebar for all users. A feature flag status mismatch had hidden it on production.
Wizard Improvements
Character face generation now succeeds for all 6 slots with automatic retry on provider failure. Voice selection works through unlimited regenerate clicks. Preset style images load correctly.
- v3.6 May 13, 2026
Report Bad Quality + Auto-Refund
Studio images now have a "Report Bad Quality" button. Submit feedback, get your credits back automatically, and optionally regenerate at no extra cost.
Lire la suite Réduire
Report, refund, regenerate
Open any Studio image in the lightbox and click "Report Bad Quality." Select what went wrong (face distortion, wrong pose, lighting issues, and more), add optional feedback, and submit. Credits are refunded immediately. Check "Try again with adjusted prompt" to get a new generation that avoids the reported issues.
- v3.5 May 13, 2026
Face Consistency Upgrade
Studio now uses native identity injection for characters with a reference photo, eliminating the painted-mask artifact on non-frontal poses.
Lire la suite Réduire
No more mask artifact
When your character has a reference photo and no custom LoRA, Studio now routes to PuLID (native Flux-based identity injection) instead of the post-generation face swap step. The result is sharper, more natural face consistency on all poses including non-frontal angles.
- v3.4 May 11, 2026
Character Link Pages
Each AI character now gets a shareable link-in-bio page. Ban-resistant, themed, and analytics-powered.
Lire la suite Réduire
Your characters deserve a home
Linktree regularly bans AI influencer accounts. RYLA now hosts your link-in-bio natively at app.ryla.ai/l/[slug]. Add your social profiles, fan platforms, and more.
Five themes, one click
Choose from gradient-purple, gradient-pink, dark-neon, minimal-light, or midnight. Your page looks premium out of the box.
Know what gets clicked
Every link click is tracked. See daily view and click counts in your dashboard analytics tab.
- v0.9 Apr 28, 2026
Quality Pipeline + Video Animation
Generate videos from portraits with Talking Head. Quality gates prevent wasted credits. Faster outfit edits via Flux.
Lire la suite Réduire
Quality-aware generation
Eye and hand quality gates validate content before credits are deducted. Character DNA extraction stores precise face embeddings. Prevents wasted credits on blurry eyes or deformed hands.
Video export from portraits
Convert any Studio portrait to a short-form video. Draft tier produces ~5-second clips for 200 credits; HQ tier for 300 credits. Downloads directly as MP4.
Talking head animation
Provide a portrait and audio URL to generate a talking head video with natural lip-sync. Draft and HQ tiers available. Track progress in a Studio modal.
Fast outfit edits with Flux
Studio edits now use Flux for standard modifications, completing in 2.4 seconds. Inpaint operations fall back to Qwen for precision masking.
Precise outfit segmentation
SAM2 and GroundingDINO masks isolate outfit regions. Flux inpaint swaps clothing without touching face or background.
- v0.8 Apr 28, 2026
AI Video Generation + Face Swap
Studio-quality 5-second video from a single photo. ReActor face swap runs automatically after generation.
Lire la suite Réduire
Text-to-video generation
Describe the scene and your character moves. Built on WAN 2.2 with 121-frame output at 24fps. Runs on H100 GPU for quality that holds at full screen.
ReActor face swap
One click maps your character's face onto the generated video. No manual masking, no keyframe alignment. Two retry attempts are built in to handle occasional GPU timeouts.
Export ready
Download H.264 MP4 directly from the studio. Optimised for Instagram Reels, TikTok, and YouTube Shorts. 960x544 at 24fps.
- v0.7 Apr 14, 2026
Character Studio Redesign
Rebuilt from scratch. New layout, faster generation, and a live preview panel so you see changes before committing credits.
Lire la suite Réduire
Live preview panel
Every parameter change in the studio now updates a live preview. No more blind generation. See what your prompt will produce before spending credits.
- v0.6 Mar 30, 2026
Outfit Presets + LoRA Training
Save outfit combinations as reusable presets. Train a custom LoRA on your character for hyper-consistent looks.
Lire la suite Réduire
Custom LoRA training
Upload 10 to 20 reference images of your character and train a LoRA in about 20 minutes. Once trained, your character stays consistent across every outfit and scene.
- v0.4 Mar 1, 2026
Studio Quality Tiers
Three generation modes: Fast for drafts, Standard for balanced speed and quality, and Quality for maximum detail. Face detection gates the Quality tier.
Lire la suite Réduire
Fast, Standard, and Quality
Pick the tier that matches your workflow. Fast mode produces draft previews in seconds. Quality mode runs a full pipeline pass for publishable results. Face detection runs automatically and blocks Quality tier when no face is found, preventing wasted credits.
- v0.3 Feb 15, 2026
Explore Gallery + Influencers Page
Browse public generations from across the platform and discover characters created by other users.
Lire la suite Réduire
Explore feed
A scrollable feed of public generations updated in real time. Filter by style, character type, or most recent. Tap any image to see the prompt and character details.
Influencers directory
A searchable directory of every public character on the platform. Each listing shows the character's style tags, creator, and recent output.
- v0.2 Feb 1, 2026
RYLA Assistant
An AI assistant in the Studio sidebar that surfaces tips, interprets feedback, and suggests next steps based on your character.
Lire la suite Réduire
Quick-start prompts
The assistant pre-loads your character context and suggests prompts tailored to your style settings. New users get from zero to first generation faster.
- v0.1 Jan 15, 2026
Identity Controls + Character Age
Characters now have an age field, a pre-prompt preview, and conflict detection for inconsistent style combinations.
Lire la suite Réduire
Age field
Set a numeric age for each character. The value influences downstream generation, and is stored per character and shown in the character header.
Pre-prompt preview
See the assembled final prompt before committing to generation. Style modifiers, character traits, and your input are shown together so there are no surprises.
Conflict detection
When age and style modifiers produce contradictory instructions, a warning appears inline. Prevents common mistakes before they cost credits.