Meta unveils new AI tools to edit images and generate videos from text instructions, based on Emu, its model for image generation announced in September 2023
Artificial intelligence researchers from Meta Platforms Inc. said today they have made significant advances in AI-powered image and video generation.
SiliconANGLEMike Wheatley
Context & Ripple Effects
Meta had already outlined a text-to-video research effort through Make-A-Video, while its September product updates put AI-generated stickers and image-editing features into its social apps. This announcement connects those parallel tracks through Emu, Meta's image-generation model.
The significance is less a single feature launch than a broader creative stack: image creation, editing and video generation can increasingly be developed from a shared model foundation. Meta had also begun applying generative image tools to advertiser workflows.
First-order effects
Meta expands the set of visual-generation tasks its Emu-based tooling can address, adding text-directed video generation alongside image editing and generation.
Meta's product and advertising teams gain a more coherent technical base for building generative creative features, rather than treating each visual format as a separate capability.
Second-order effects
The move raises pressure on other consumer platforms and creative-tool vendors to pair generation with editing controls, not merely offer one-shot image or video output.
Meta's earlier AI creative tools for advertisers provide a clear adjacent route for these capabilities: faster production of campaign variants and visual assets if they are incorporated into ad products.
Third-order effects
If these tools move from demonstrations into Meta's high-reach products, generative visual creation is likely to become an embedded platform function, with distribution and workflow integration mattering alongside model quality.
The combination of synthetic images, edits and video will increase the importance of product-level controls for how generated media is labeled, managed and used across social and advertising surfaces.
The trend: This is one data point in the shift from standalone generative-media models toward AI creative tooling embedded in large consumer and advertising platforms.
Yeah it plays fast and loose with the rules. But they spelled some people's names wrong and don't understand nicknames. So they block Sarah Anderson instead of Andersen. They block Hulk Hogan but not Terry Bollea nor Hulkamania. They don't block OJ Simpson because I guess he …
Today we're sharing two new advances in our generative AI research: Emu Video & Emu Edit. Details ➡️ https://ai.meta.com/... These new models deliver exciting results in high quality, diffusion-based text-to-video generation & controlled image editing w/ text instructions. 🧵 [vid…
@AIatMeta Interesting and hope the documentation is a great as others. This will continue to push the boundaries of what is possible and raising tide rises all pursuits.
Emu now welcomes the Edit and Video variants: amazing bleeding edge research in diffusion models by @Andrew__Brown__, @imisra_, @deviparikh and team! I hope the internal tides move towards having this open 🌊 @ylecun #freeEmu! 😊 Emu Edit and Video 📝: https://ai.meta.com/...... [im…
@multimodalart ... Speaks to how awesome @ylecun and @AIatMeta have been that I was surprised it wasn't open 😅 My guess, it's about the data. These both seem to be very data hungry capabilities & who knows where it was sourced
Meta Research is working on Emu Video — an AI video model intended to rival the likes of Pika and Runway. They compared results with human raters and concluded their “model outperforms commercial solutions such as RunwayML's Gen2 and Pika Labs.” Wdyt? [video]
@AIatMeta Emu edit not publicly available? You can do this same flow in Cauldron a custom GPT built in 2 days https://chat.openai.com/... https://x.com/...
Meta's Emu: Is It Curtains for Photoshop? Meta's latest AI toys, Emu Video and Emu Edit, are here to show Photoshop how the cool kids edit now. Emu Edit lets you boss around your photos with just text. “Make the sky purple, add a unicorn”—bam, done! Who needs tedious clicking... …
Depth mask is a little questionable. I think thats just an greyscale invert, but I did get the rough hedgehhog shape And keep the surprised look [image]
Real cute Meta no public release?! sigh... Luckily for us This same flow is available today Using Cauldron: A custom GPT built for image mixing [image]
Cauldron did really struggle with the hands But did manage to find “Hands” aka rain and spacesuit patches Technically correct on the last one after adding gloves lol [image]
I thoroughly enjoy Meta's naming scheme for its AI projects. Calling them things like Llama and Emu makes them feel less like Skynet and more like something innocent and fun.
10 years ago adobe's pixeltone concept [ https://www.youtube.com/...] envisioned a natural language interface for photo editing but it never really went anywhere but here we are today with meta's emu edit!! exciting :) (why'd they remove the poodle tho 😭) [video]
2️⃣ Emu Edit This new model is capable of free-form editing through text instructions. Emu Edit precisely follows instructions and ensures only specified elements of the input image are edited while leaving areas unrelated to instruction untouched. This enables more powerful... […
So many announcements today. Meta just dropped EmuVideo, generating 4-second short videos at 512x512 resolution and 16 FPS. Idea is quite straightforward: text -> image first, then do a “super-resolution” of the image along the temporal axis to synthesize motion. Long-form... [vi…
Ok, they did get segmentation much better than Cauldron could I wonder if I could import something into code interpreter that would do this better. Hmmm [image]
The best thing about this was they've transformed image editing to talking to the AI. So many image editing softwares are just a mess of features that few know how to use.
Building on our work of a strong image foundation model, Emu, excited to announce Emu Video and Emu Edit from our group @AIatMeta. The potential of these models to enable more expression on our platforms is tremendous and am excited for 2024! https://ai.meta.com/... [video]