Changelog

We've shipped a lot since the last update: a new way to explore ideas without losing your thread, document uploads, clearer credit spending, and a fresh wave of image, video, and audio models.

Branch your chats

Ever wanted to try a different direction without throwing away the conversation that got you there? Now you can. Branch any chat from a completed assistant turn and pick up from that exact point in a brand-new conversation.

• Hover over any finished turn and hit Branch to fork the chat — every message, asset, and bit of context up to that point comes with it.
• Your original chat stays untouched, so you can explore a wild idea, compare two approaches, or back up and take a different path whenever a generation goes sideways.

It's the fastest way to say "what if?" without starting over.

Bring your documents into chat

Glif now understands documents, not just media. Attach or paste .pdf, .txt, and .md files straight into the composer and the agent can read, summarize, and work from them.

• Drop a PDF and ask for a summary, key points, or a rewrite.
• Paste a document right into the composer — it's picked up automatically.
• We also raised the audio upload limit to 20 MB so larger clips come along for the ride.

See exactly what you're spending

Credits got a lot more transparent.

• Per-chat credit totals show how much a conversation has cost at a glance.
• Per-turn cost breakdowns explain what each step spent and why.
• Credit details are hidden by default and one toggle away whenever you want them — and estimates are rounded so they're easier to read.

New models

A big batch of new and upgraded models landed across every modality.

Image

• Nano Banana Pro — now running on Google's latest stack: faster, with up to 4K output and up to 14 reference images for strong multi-subject composition. Great for character composites, product shots, posters, and edits that combine many references.
• MAI Image 2.5 — generate from a text prompt, or pass a reference image to edit it. Good for illustrations, photos, concept art, product shots, and prompt-driven edits.

Video

• Veo 3.1 — animate any image into video with natural audio, now on Google's latest API for our highest-quality Veo image-to-video.
• Seedance 2.0 Mini — fast video with synchronized native audio and tight lip-sync, from text or an image. Supports start-and-end-frame interpolation and clips up to 15 seconds.
• Kling V3 Motion Control — transfer movement from a reference video onto a character image for motion transfer, dance, and gestures. Toggle standard for value or pro for top quality.
• Hyperframes — render programmatic, frame-perfect videos from an HTML composition, with precise timeline control. Perfect for animated titles, lower-thirds, data-driven motion graphics, and overlays.

Music & audio

• Lyria 3 (Google DeepMind) — compose original music from a text prompt: describe style, mood, BPM, key, instruments, and song structure, drop in lyrics for vocals, and generate anything from a short clip to a full-length track.
• Seed Audio 1.0 — text-to-speech with 20 preset multilingual voices (English, Chinese, Japanese, Spanish, and more) plus voice cloning from up to 3 reference clips. Great for voiceover, narration, dialogue, and character voices.

We also retired a batch of underperforming image and video tools so the agent picks the right one for the job more often.

Also new

• Clearer tool approvals — pending and in-progress tool calls now show up in the approval tray and your chat library, so you always know what the agent is doing.
• Connected apps — review and revoke the third-party apps you've connected to Glif from the API tokens settings page.
• A smoother, faster app — a wider sidebar, virtualized chat lists for snappier scrolling, polished mobile media panels, message queuing while uploads finish, and composer prefill from the homepage so you can start a chat in one tap.

More on the way. Drop into our Discord and tell us what you'd like to see next!

Choose your agent's intelligence level

You can now pick the brain that fits the task, right from the composer. Four levels, four trade-offs between speed, smarts, and cost:

• Glif Lite ($) — our most affordable model, capable and easy on credits for high-volume, everyday work.
• Glif Flash ($$) — fast and great for tasks with a human in the loop.
• Glif Smart ($$$) — a balanced model that combines strong reasoning with speed for everyday complex tasks.
• Glif Genius ($$$$) — our most capable model, best for autonomous tasks that require the highest level of intelligence.

The agent now also consumes credits that scale with the complexity of your requests — simpler asks cost less, harder ones cost more. Reach for Lite or Flash when you're iterating quickly alongside the agent, and switch up to Genius when you want to hand off something hard and let it run.

New models

We have a host of new models available in Glif:

• Krea 2 — high-quality image generation from a text prompt, with up to 10 style reference images. Great for photoreal images, concept art, illustrations, product shots, and posters. Pick the large variant for top quality or medium for half the cost.
• Grok Imagine 1.5 — animate any image into video with native audio. Supports 480p or 720p and 1–15 second clips. Perfect for motion, cinematic shots, b-roll, and short-form video.
• Phota — an image editor for personalized, identity-preserving edits. Combine up to 10 reference images for character composites, portrait restyling, and photo retouching.
• P-Image Upscale — the fastest upscaler we offer (under a second), with outputs up to 128 MP. Increase resolution, enhance textures, and restore detail, with optional enhancement tuned for AI-generated images.
• MiniMax Music 2.6 — generate complete tracks with vocals, backing music, and detailed arrangements from lyrics and a style description.

Also new

• Clearer tool approvals — pending tool approvals now show up in the composer tray and even in your browser tab title, so you'll never miss one waiting on you.
• A snappier composer & library — new chats autofocus the composer, chat example cards got a cleaner look, and the library and chat detail views are smoother on mobile.

More on the way. Drop into our Discord and tell us what you'd like to see next!

Create your own skills

A skill is a reusable bit of Glif: a workflow, a persona, a recipe, a style — anything you'd otherwise have to re-explain at the start of every chat. Glif already comes pre-loaded with a bunch of really useful skills built by our team, but now you can build your own skills too!

The best way to make one is the way you already use Glif: just talk to it.

• Build skills in chat. Once you've had a chat that does something useful, ask Glif to turn it into a skill. It'll write the skill for you and save it. Works in any conversation.
• Find them under Customize. Every skill you create lives in Customize in the nav, ready to edit, share, or invoke.
• Invoke a skill by typing / in any chat. The composer pops up a skill picker; pick one and the agent kicks off with that skill loaded.
• Tell Glif about yourself. Ask Glif to interview you about your work, your style, or your projects — or just say "remember that I…" and it'll commit the fact to user memory for future chats.

Every chat now starts from a Glif that knows what you're working on, what tools you like, and how you want it to respond. The difference between a powerful generic agent and your agent.

A smarter library

Finding the media you upload to Glif — or make with it — got easier.

• All media is searchable via ⌘K — jump to any image, video, or audio file from anywhere in the app.
• Tabs and filters in the Library — uploads and generations are now broken out into their own tabs, with inline search and filters to narrow down further.
• Mention assets in new chats with @ — type @ in the chat composer to pull a specific file from your library straight into the conversation as a reference.
• In-chat media sidebar — a new sidebar surfaces media relevant to the current chat as it is generated, so you can browse what's already in play without leaving the conversation.

New native tools

We are constantly adding new capabilities to the Glif agent. Here are a few of the latest:

• VEED Remove Background (Video) — strip the background out of any video, not just stills.
• ElevenLabs Scribe v2 — upgraded speech-to-text with better accuracy and language support.
• Librosa Beat Analysis — extract tempo, beats, and rhythm structure from audio. Useful for syncing video cuts to music, scoring, or any beat-aware composition.
• Kling 4K — generate video at 4K resolution.

We also updated tool descriptions to make the agent better at choosing the right tool for the right job, along with a host of smaller UX improvements and bug fixes.

More on the way. Drop into our Discord and tell us what you'd like to see from Glif!

Today's the day: Glif 2.0 is officially launching. And because there's no better way to explain what Glif can do than to let Glif do it, we asked the agent to make our launch videos for us. It made five.

Glif is a creative super agent: tell it what you want to make and it produces incredible outputs using virtually every available AI model. Create ads, marketing content, films, short-form content, voiceovers, music, and more — all in one conversation. Easy to start, endlessly deep.

Creatives: you're not cooked. You're the chef now.

We raised $17.5M to build this

We're also announcing our $17.5M seed round, led by a16z and USV, with participation from Interface, Two Small Fish Ventures, and SpiceCap. Huge thanks to our investors, our community, and everyone who's been making wild things alongside us.

Five launch videos, all made in Glif

Every one of these was created in a chat session at glif.app — no editing suite, no timeline, no compositor. Different prompts, different models, different vibes, same agent.

1. "You're not cooked. You're the chef now."

The hero video. Our case for creatives in the age of AI.

2. "It's time to vibe"

An anime brief handed to Glif. It reached for GPT Image 2 and Gemini 3.1 for stills, Seedance 2.0 for motion, and Google's new TTS for the protagonist's voice. Our 8-year-old's verdict: "first good AI thing you've made."

3. The AI influencer at a whiteboard

Glif flexed its AI influencer skills: an influencer at a whiteboard explaining Glif with a new scribble in every scene. Stills from GPT Image 2, video from Kling o3, plus subtitles and music. Favorite moment: when the whiteboard starts shaking.

4. The Interviews

Extremely cheesy, extremely funny. We asked Glif for portraits of people on black backgrounds, iterated on the script with it, and it animated the whole thing using a mix of Veo and Kling.

5. "Cooked"

A second pass on our original "cooked" script — same idea, totally different setting and vibe. Glif leaned on Kling o3's transitions throughout, with stills from GPT Image 2 and Gemini, and voiceover from ElevenLabs.

Go make something

All of the above was made in chat sessions at glif.app. Start a conversation, describe what you want to make, and let Glif pick the tools.

We can't wait to see what you cook up. 🍳

Since we launched Glif 2.0 a few weeks ago, we've been heads-down shipping. Here's a roundup of the public-facing changes.

Queue up messages while the agent works

You no longer have to wait for the agent to finish before typing your next instruction. Hit enter while a turn is in progress and your message joins a queue that's sent as soon as the agent's ready. You can edit or cancel queued messages before they run, and there's an indicator above the composer when you're editing one in place.

New native tools

A big batch of new agent capabilities across image, video, and audio:

Images

• Z Image (with LoRA + ControlNet support) — this model punches way above its weight class. Fast, cheap, very good. Supports text-to-image, image-to-image, and ControlNet modes with optional LoRA.
• Quiver — generate real SVGs from text or reference images. Clean, production-ready vector output, infinitely scalable. Great for logos, icons, and illustrations.
• HTML to Image — have the agent write HTML + CSS and get back a rendered PNG. Ultra-useful for social cards, posters, infographics, and any layout you want repeatable and pixel-perfect.

Video

• PixVerse V6 — a strong all-rounder that's especially good at action scenes. Image-to-video with native audio (background music + SFX, not speech), plus style presets (anime, 3D animation, clay, comic, cyberpunk). Also available as a keyframe transition model.
• Seedance 2.0 — ByteDance's latest text/image-to-video model with native audio, in four variants (standard, Fast, Reference, Fast Reference).
• Remotion — have the agent write a React component and get back a fully rendered video. Programmatic, repeatable, extremely powerful and versatile.
• Fabric 1.0 LipSync — fast, clean lip-sync from a portrait image and an audio clip. Another solid lip-sync option in the arsenal.
• Remove Video Watermark — strips watermarks from videos via WaveSpeed, billed by duration (clips up to 10 minutes).

Audio

• Google Gemini TTS — premium-quality text-to-speech powered by Gemini 3.1 Flash. 30 voices, 70+ languages (auto-detected), and inline audio tags for expressive control like [whispers], [excited], [laughs], and [sighs].

A better media library

Your library is easier to navigate:

• Group by chat — the library now groups media by the chat that created it (this is the default), so you can jump back to the source conversation. Group-by-date and ungrouped views are still available.
• Like anything — hit the heart on any output in chat or in the library to save your favorites, then flip the "Liked only" toggle in the library to filter down to them.

General UI improvements

A broad polish pass across the app, plus a bunch of smaller fixes:

• Refreshed dark mode with a cooler palette
• Cleaner settings pages
• More reliable video player
• Videos in galleries now display at their true aspect ratio
• Attached media inside chats lays out as a clean grid
• Tidier chat layout — task groups, code blocks, and inline forms
• Better markdown rendering, including tables, blockquotes, and tool-task thumbnails
• Refreshed favicon

More coming soon. As always, drop into our Discord and let us know what you're making (or what's broken).