Midjourney just launched Omni-Reference, a new feature built to drop specific characters, objects, and vehicles directly into generated scenes. It basically acts as a "put THIS in my image" command, giving creators tighter consistency control without endlessly rerolling prompts.
Higgsfield just introduced a new Start & End Frames feature that lets users upload both an opening and closing image to shape their video. Instead of guessing where the camera will wander, creators can now enforce purposeful motion and a clear visual trajectory between two key points.
Ideogram just rolled out a major update to Ideogram 3.0, boosting realism, style versatility, and prompt adherence. Creators can now use Magic Fill and Extend inside Ideogram Canvas to edit uploaded or generated images, and the 3.0 model is officially live on the developer API.
Suno just launched v4.5 for Pro and Premier subscribers, delivering punchier vocals, detailed audio textures, and seamless genre blending. The standout update is length: you can now generate full tracks up to 8 minutes long in a single run without using the extend tool.
Pika just rolled out new hairstyle effects for its Pikaffects tool across the web and iOS. You can now turn anyone's hair pink, stretch it ridiculously long, or shave it completely bald with a click. It is a fun, accessible way to spin up stylized video transformations in seconds.
Krea AI just launched GPT Paint, a new feature inside Krea Image that lets users visually prompt ChatGPT. Instead of relying purely on text descriptions, creators can now guide the AI using edit marks, basic geometric shapes, notes, and reference images. Why it matters: It bridges the gap between rough sketches and prompt engineering, giving creators far more precise spatial control.
Midjourney just rolled out an update to its V7 model, boosting overall quality alongside hand and body coherence without requiring new settings. The release introduces a redesigned lightbox editor with smart segmentation, plus a new experimental --exp aesthetic parameter that adds punchy detailing and tone mapping to your generations.
Runway just rolled out Gen-4 References to all paid users, tackling one of generative video's biggest headaches. Creators can now feed photos, generated stills, 3D models, or selfies into the model to keep characters and environments consistent across scenes without prompt wrestling.
Freepik just released F Lite, a 10B-parameter open image generator built with Fal.ai and trained entirely on 80 million licensed, safe-for-work images. Designed to sidestep copyright headaches, it comes in standard and texture flavors, though running it locally requires a beefy GPU with at least 24GB of VRAM.
Natasha Lyonne is set to direct and star in Uncanny Valley, an upcoming sci-fi feature that will use Moonvalley's text-to-video model Marey for visual elements. Moonvalley claims the model is trained entirely on licensed content with compensated creators, aiming to bypass Hollywood's copyright backlash. The film currently lacks a release timeline.
ShengShu Technology just launched Vidu Q1, an end-to-end video model that pairs cinematic visuals with synchronized 48 kHz audio. It introduces a First-to-Last Frame feature to generate smooth transitions between unrelated images in 1080p, while matching audio to emotional tone. AI video is inching closer to full-fledged Hollywood post-production.
Luma AI just dropped the Ray2 Camera Concepts API, giving developers granular control over cinematic framing and camera movement. It rolls out starting with Motion and Angles concepts, bringing AI video generation one step closer to actual film directing.
Higgsfield AI just launched Iconic Scenes, letting creators drop a single selfie straight into legendary Hollywood moments. Users can pick from over 20 templates like The Matrix and Interstellar, complete with 80+ cinematic camera controls and AI style presets. It basically brings instant, zero-crew movie cameos to anyone with an internet connection.
Tavus just rolled out Hummingbird-0 on fal as a research preview, touting it as the most accurate zero-shot lipsync model they have tested across open and closed source. The release aims to deliver significantly cleaner, more convincing video dubbing and talking heads without extensive fine-tuning.
Argil just merged its models with OpenAI tech to allow AI avatars to physically hold products while speaking on camera. The entire showcase video can be generated directly from a single product image, making e-commerce ad generation far more convincing.
Krea AI just integrated ChatGPT's image capabilities into its platform. Creators can now edit images, render 3D scenes, and swap between over 300 styles through simple chat prompts in under 30 seconds. It brings conversational iteration directly into creative design workflows.
Adobe just launched its next-gen Firefly Image Model 4 and 4 Ultra, bringing 2K resolution, sharper text generation, and style matching. It also opened its 1080p video model to everyone and revamped the Firefly web app—which now surprisingly lets users switch to models from OpenAI, Google, and Flux.
Grammy-winning artist Imogen Heap just rolled out five official "stylefilters" on AI music platform Jen, letting users generate tracks based on her catalog starting at $4.99. Unlike rivals battling major copyright lawsuits, Jen strictly trains on licensed audio. Why it matters: It offers a rare, artist-endorsed blueprint for ethical AI music monetization, though creations are limited to instrumentals for now.
Higgsfield AI just launched Turbo, its fastest video generation model to date. The update speeds up rendering by 1.5x, cuts costs by 30%, and adds priority queue access alongside seven new motion styles. It makes rapid, iterative video experiments far cheaper and faster to run.
Tencent just upgraded its Hunyuan 3D AI Creation Engine to v2.5, bumping the geometric model from 1B to 10B parameters with 10x more mesh faces. It introduces multi-view PBR generation, automatic bone binding, and doubles daily free generation quota to 20.
OpenAI just launched its gpt-image-1 model in the API, giving developers direct access to the native multimodal engine powering ChatGPT. The model excels at accurate text rendering and custom styling, with token-based pricing translating to roughly $0.02 to $0.19 per square image.
Character.AI just unveiled AvatarFX, a video generation model that brings static images to life with synchronized speech, singing, and body gestures driven by audio. Built on DiT architecture with distillation for fast inference, it supports multiple speakers and will roll out first to CAI+ subscribers in the coming months.
Dreamina AI just launched Seedream 3.0, its latest image generation model featuring native 2K resolution. The upgrade focuses on cinematic visual quality, realistic textures, and cleaner typography with accurate English text rendering. It is now live for users to test on the web.
Krea just launched Krea Stage, an experimental tool that builds fully editable 3D environments from a single 2D image or prompt. Creators can adjust lighting, shift perspectives, and generate camera-consistent videos without wrestling with complex 3D software. Why it matters: It completely bypasses the brutal learning curve of traditional 3D modeling.
Arcads just launched Gestures, letting creators prompt unlimited emotions and movements for their AI avatars. Users can simply pick an avatar and type an emotion to instantly generate dynamic video hooks, solving the stiff-puppet problem that plagues many AI video ads. It is now live for all Arcads users.
LTX Studio just rolled out a major update featuring a built-in timeline editor and custom AI actor training. Creators can now trim clips, reorder sequences, and sync audio directly in the workspace, while locking in consistent faces and outfits from reference images. It turns the AI filmmaking platform into a much more cohesive post-production suite.
Kuaishou just unveiled Kling AI 2.0 alongside its new Kolors 2.0 image model, topping over 22 million global users. The flagship upgrade introduces a Multi-Elements Editor that lets creators add, remove, or swap objects directly inside existing video clips using text and image prompts. Granular video control is finally catching up to static image editing.
Google just opened its Veo 2 video model to developers through Google AI Studio and the paid Gemini API tier. The model turns text or image prompts into eight-second clips with realistic physics and camera controls. It means apps can now generate commercial-grade motion clips without custom video pipelines.
YouTube just launched an AI feature inside Creator Music that lets creators generate custom instrumental backing tracks from text prompts. Eligible creators can open the "Music Assistant" tab to specify instruments or moods, producing tracks that are free to use without copyright claims. It is gradually rolling out to U.S. creators in the YouTube Partner Program.
Google just revealed how it used AI to upscale the 86-year-old film The Wizard of Oz for the Las Vegas Sphere's massive 16,000x16,000 screen. Using fine-tuned Gemini models, Veo 2, and Imagen 3 alongside studio archives, the team expanded backgrounds and digitally placed off-camera characters into widescreen scenes. With over 90% of the film touched by AI, it debuts on August 28th, 2025.
Amazon just upgraded its Nova Reel AI video model to version 1.1, doubling max clip duration to two minutes. The update adds a new Multi-shot feature that keeps visual styles consistent across sequences, plus a manual mode to direct up to 20 shots using a single reference image and prompt.
