Magic Hour Changelog: 2026

Magic Hour Video is now available in the Framer Marketplace. Select an image in your design, describe its motion, preview the generated clip, and add it as a native Framer Video component without leaving the editor.

Use existing artwork to create motion for your site while keeping the original image in place. The plugin is free; generation requires a Magic Hour account, API key, and sufficient API credits.

For example, select a product image, generate a short motion clip, then choose Add video to design. The inserted video uses muted looping playback; a seamless visual loop is not guaranteed.

Browse 100 meme formats in the new Meme Templates directory. Each format has its own page with an explanation, an example idea, captions, and the matching generator already selected.

Find a format that fits your joke before writing the idea, without searching through the generator's full template list.

For example, open Drake Hotline Bling to see how its reject-and-approve layout works, then enter your own idea in the selected template's generator.

The dashboard's AI Voice Generator, Voice Changer, and Voice Cloner now include an expanded catalog of curated voice presets. New Voice Generator and Voice Changer sessions start with Ashley selected.

Explore more voices directly in the tools when choosing an identity for narration or a voice workflow.

For example, open the voice picker and compare Ashley, Wade, Lauren, or Nicholas before making your selection. The expanded dashboard catalog is separate from the public Text to Speech page's voice selection.

Generate Script is now available in the signed-in UGC Ad Generator. Choose an actor or opening image, then generate a script or revise an existing draft with your selected duration in mind.

Develop the words alongside the actor and product instead of moving between writing and video tools. You can regenerate the script or undo the change before creating a video.

For example, select your actor, add a short product idea or draft, and choose Generate Script. Review the returned text before generating. This works in text-script mode, not uploaded-audio mode.

Video Upscaler is now available through the API for paid plans. Choose a target resolution of 720p, 1080p, 2K, or 4K while preserving the source video's aspect ratio.

Upscale clips inside your own application or video-processing workflow without manually moving between tools. Processing uses Magic Hour credits and supports up to 4,000 frames at a maximum of 30 fps.

For example, upload a 720p clip and request a five-second segment at 1080p with POST /v1/ai-video-upscaler. Use the returned project ID to check completion and retrieve the download. Choose a target above your source resolution: a clip already at the target resolution may be returned unchanged and still use credits.

The Music Generator API is now available. Describe a track, choose a duration from 15 to 300 seconds, and request instrumental music or provide lyrics.

Add music generation to your own app or automated video workflow using your Magic Hour API key and credits.

For example, request a 30-second instrumental track with the prompt "Warm acoustic guitar for a product walkthrough." Submit it to POST /v1/ai-music-generator, then check GET /v1/audio-projects/{id} for completion and download links.

Video Expander now prices new generations at 24 credits per second for 480p, 48 for 720p, and 72 for 1080p. Previously, all three resolutions used the 480p rate.

Higher-resolution expansions now cost more credits. Auto uses the resolved output resolution. Existing jobs are not repriced, and resolution access by plan is unchanged.

For a five-second clip, that means 120 credits at 480p, 240 at 720p, or 360 at 1080p. Review the displayed credit cost before generating.

Video Colorizer now supports clips up to five minutes in the full creation flow, up from 45 seconds. The free-tool limit remains 10 seconds.

Colorize longer scenes without first splitting them into short clips.

For example, submit a one-minute black-and-white clip as a single Colorizer job, then download the completed video.

Image Body Swap now lets you choose between Nano Banana and FLUX.2 Klein. Nano Banana remains the default.

Choose a model and supported output resolution to fit your body-swap workflow. FLUX.2 Klein offers 480p, 720p, 1080p, and 1440p settings; credit cost depends on the resolution.

For example, choose FLUX.2 Klein with 720p for an image body swap, or select 1080p when you need a larger output. Review the displayed credit cost before generating.

MiniMax H3 is now available in Keyframes mode. Add multiple images to guide a video through an ordered sequence, rather than providing only its first and last frames.

Set the sequence and segment durations while H3 prepares the transition prompts. This workflow does not offer editable per-keyframe prompts.

Example: arrange three images as two transitions and generate the sequence.

Saved Characters are now available in Storyboard's Add Image flow and through explicit Character selection for UGC Actor, Clothes Changer's Person Image, and Head Swap's Head input.

Reuse a saved identity in these inputs without finding and uploading its image again. The selected Character fills the relevant input; you can review the setup before generating.

Example: choose a saved Character as your UGC Actor and review its image in the Actor input and Opening Shot.

Completed Video Upscaler results now support a before/after slider and a side-by-side comparison view.

Inspect the result against the original without opening separate players. Comparison is available when the selected trim begins at the start of the source video, keeping both sides synchronized.

Example: open a supported completed upscale and switch between the comparison slider and side-by-side view.

Character creation and editing now let you choose a saved voice, record a voice, or upload voice audio without a preview setting. The same controls are available when adding a Character from AI Avatar.

Keep a reusable visual identity and voice together instead of setting up the voice again for each Character. New celebrity-preset selection is removed; existing Characters keep their previously saved voice when you leave it unchanged.

Example: edit a Character, choose a voice from My Voices or record one, then save.

Seedance 2.0 Fast is now a distinct option in Text-to-Video, Image-to-Video, and the public API. It is available on paid tiers at 480p or 720p for clips from 4 to 15 seconds.

Choose Fast explicitly rather than treating it as an interchangeable fallback. Supported workflows include audio, end frames, and media references; review the displayed credit quote for your settings.

Example: choose Seedance 2.0 Fast in the model picker and set your clip length and resolution. Release checks verified its model selection and request routing. The first-party Seedance family showcase below is not a measured Fast speed or output-quality comparison.

Completed dashboard video outputs now offer Extend Video and Expand Video in their menus, including supported Library detail views and mobile result menus.

Open the next tool with that output already loaded instead of downloading and uploading it again. Choose your trim, aspect ratio, and other settings before clicking Generate; opening the shortcut does not start a generation.

Example: open a completed video's menu, choose Expand Video, and review the prefilled clip. Release checks verified this handoff on desktop and mobile. The first-party illustration below shows Video Expander, not the menu. Image and audio output menus are unchanged.

The AI Talking Photo API now accepts an explicit resolution value of 480p or 720p. The older max_resolution field is deprecated but remains compatible.

Specify the output resolution directly in new integrations. When both fields are supplied, resolution takes precedence; supported model and plan limits still apply.

Example: set resolution to 720p in a new request. Existing integrations can keep using max_resolution while migrating.

The public API now includes POST /v1/ai-ugc-ad. Supply a prepared image in assets.image_file_path and a script to create a UGC video using supported Kling 3.0, LTX 2.5, or Veo 3.1 options.

Integrate this prepared-image workflow without generating and polling a separate opening image. The endpoint returns a standard video project ID and uses the existing status and download endpoints; plan, duration, resolution, and credit limits still apply.

Example: send a prepared image and script to POST /v1/ai-ugc-ad, then use the returned project ID to check status and download the video.

MiniMax H3 now accepts both a first and last image in Image-to-Video at 480p, 720p, and 1080p.

Choose where the clip begins and ends instead of guiding it with only a starting frame. Add a motion prompt to describe the transition between the two images.

Example: select MiniMax H3 in Start & End Frame mode and add both images. Release checks verified the uploads at 480p and completed worker output checks. The existing first-party H3 example below is not that test output.

Video Upscaler v2 is now available in Magic Hour, with a short free preview on the public tool page and credit-based 2x upscaling available to Free-plan accounts.

Try the first three seconds before committing to a longer upscale. The free preview uses 2x scaling, is capped at 2K, and has a three-per-day allowance; longer generations use credits and applicable limits.

Example: upload a clip and review its short upscale preview before choosing a longer run. The existing first-party example below illustrates video upscaling, not a measured v2 quality or speed comparison.

The public Talking Photo and Lip Sync audio pickers now include Record audio alongside uploads, text-to-speech, and sample audio.

Speak your own script without making a separate audio file first. Your selected photo or video stays in place, and cancelling the recorder keeps the existing input.

Example: open Add Audio, choose Record audio, allow microphone access, and finish your recording. Release checks verified that recorded WebM audio can produce a Talking Photo result. The first-party demo below illustrates Talking Photo, not that specific recording.

Wan 3.0 video creation now sends supported image, video, and audio references directly to the model instead of first synthesizing a hidden starting image.

Guide a new clip with the media you selected. Use native references as a separate workflow from first/last-frame generation; the two input modes cannot be combined.

Example: choose Wan 3.0, add reference media, and describe the clip you want. Release checks produced completed native-reference videos through both supported providers. The first-party video illustration below is not a Wan reference-quality benchmark.

Signed-in users can now Enhance supported Text-to-Video prompts and LTX 2.5 Image-to-Video prompts without a preview setting. Image-to-Video uses the uploaded frame as context and can suggest motion from a blank prompt.

Review the suggested direction before generating, or choose Undo to restore your original wording. Existing model and sign-in restrictions still apply.

Example: upload a starting frame, choose Enhance, and review the suggested motion.

Nano Banana 2.1 is now available as a distinct model in AI Image Generator, AI Image Editor, and the public API using nano-banana-2.1.

Choose the new model for your image workflow while existing Nano Banana 2 API and template callers remain supported. Review the credit quote for your selected resolution and image count before generating.

Example: select Nano Banana 2.1 for a new image or edit, or send its model identifier in an API request. The existing first-party Nano Banana family showcase below is illustrative, not a benchmark of version 2.1.

Qwen Image 2.1 now accepts up to three primary reference photos in AI Image Generator and AI Image Editor. Free-plan users can use this supported reference workflow with credits.

Guide a creation with saved Characters or reference photos without a separate reference-feature subscription gate. Model access and credit costs still apply; Qwen Brand Kits are not supported.

Example: choose Qwen Image 2.1 and add reference photos to guide your prompt. The existing first-party Qwen showcase below illustrates image editing, not a three-reference quality benchmark.

Signed-in users can now use Enhance on supported AI Image Generator models and single-image AI Image Editor workflows without a preview setting.

Review a revised prompt before generating, and use Undo to return to your original wording. Guest, unsupported-model, multi-image, and video restrictions still apply.

Example: write a product-photo direction, choose Enhance, review the revised text, and Undo if you prefer your original.

Kling 3.0 now supports one video reference, optionally combined with up to four image references, in its supported paid reference workflow.

Guide a new clip with an existing video and supporting images. Supported reference clips can retain or mute their original sound; this does not add newly generated audio.

Example: select Kling 3.0, add a video reference and an image, and describe your clip. Release evidence includes a completed three-second 720p reference generation with original sound. The existing first-party Kling showcase below is not that test output.

Completed still-image results now offer Save as… in their menu. Name the image in the Save as Character dialog to add it to your Characters picker.

Reuse an image you already created as a Character without limiting this action to AI Influencer results. The action is not available for GIF, video, audio, processing, or failed outputs; each save creates a new Character.

Example: open a completed image's menu, choose Save as…, enter a name, and save. Release checks verified that the saved Character appears in the Image Generator picker on desktop and mobile. The first-party Characters showcase below illustrates reusable Characters.

AI Image Editor and AI Image Generator offer 4:3, 3:2, 4:5, and 2:3 alongside 16:9, 9:16, and 1:1. Auto remains available in the editor. Supported models include Qwen Edit, Flux 2 Klein, Nano Banana, FLUX Schnell, and Z-Image Turbo.

Choose an output shape that fits your photo or intended placement instead of being limited to square, portrait, and landscape presets.

Example: choose 4:5 for a portrait-oriented image or 3:2 for a wider photograph.

Explore new free image workflows for product mockups, product staging, room redesigns, flyers, invitations, and other designs. Image-editing tools accept the inputs their workflow needs; generator tools start from your subject or prompt.

Start with a task-specific tool instead of rebuilding a prompt from scratch. These tools produce image outputs; a design or presentation tool name does not imply editable slides, PDFs, or vector-file exports.

Example: upload a product image to AI Mockup Generator for a directed mockup workflow. The first-party product-photo example below appears on that tool's page and illustrates image editing, not a benchmark of the mockup preset.

AI Music Video Generator now offers Timed directions without a preview setting. Add moment-specific prompts on the timeline, then edit their timestamps and directions below the main prompt.

Guide what should happen at different points in your music video instead of describing the entire sequence in one paragraph. Save applies your moment edits; Cancel leaves the existing directions unchanged.

Example: choose a point on the timeline and write a direction for that moment, then add another moment later in the clip.

Video-to-Audio now lets you set both the start and end of the clip you want to process. The selected range is preserved when you Recreate or Retry the project.

Generate audio for the relevant moment instead of processing from the beginning of your source video. Validation and credit calculation use the selected range.

Example: release validation selected seconds 40–45 of a 45-second video and produced a five-second result with audio. The first-party product demo below illustrates the Video-to-Audio workflow, not that validation recording.

Qwen Image 2.1 is now available for image generation, image editing and the public API. It is the default model when starting a new AI Image Editor session.

Start an edit with Qwen Image 2.1 without finding a preview setting. Your explicit model selections remain respected.

Try a concrete edit such as changing a background or an outfit. The first-party examples below illustrate these editing workflows; they are not a Qwen Image 2.1 quality benchmark.