Runware

Introducing Runware Serverless. The lowest-cost way to run your own custom AI workloads. $0 while idle. Then from $0.63/GPU-hour. Deploy Python code or a container, choose your GPU and configure scaling. We handle the scheduling, queues, workers and routing.

@ByteDanceSeed_ Seedream 5.5 is live on Runware Building on 5.0 Pro, 5.5 brings transparent output to text-to-image, layer-count control and text recognition for decomposition, and multi-turn edits. ByteDance reports unedited areas stay pixel-stable across 100 rounds, a +55.62% preference for image-to-image realism, and more than double 5.0 Pro's usability in film and office work. - Set settings.background to transparent for RGBA output, from a prompt or from a reference photo with no alpha channel - Split one image into a base plus up to 99 layers, with the text, font and color returned for text layers - Edit with up to 10 reference images - 1K, 1.5K and 2K tiers, or custom sizes up to 4.6MP from 1:16 to 16:1 Already calling Seedream 5.0 Pro on Runware? Switch your model ID to bytedance:seedream@5.5 from $0.02 per image

We're giving $3K in Runware credits to the runners-up in the @HeyGen Video Contest! To participate, ask your AI agent to make a video with HeyGen Video 1.0 through HeyGen's MCP, then tweak and edit it however you want. Check out the details to get involved: heygen.com/video-contest #HeyGenVideoContest

@MiniMax_AI H3 Image & H3 Image Highspeed are live on Runware on D0 MiniMax built its new image model on H3, its popular open-weight video model, calling it the strongest open-weight image model released so far. - Edit with up to 16 reference images - Output up to 20 MP, with 1K, 2K and 4K presets or exact width and height - Aspect ratios from 1:3 to 3:1 - H3 Image Highspeed runs at 2x the speed, for quick edits and fast iterations Image output from $3.80 per million tokens, with 50% off 4K outputs on both models through November 17

@PrunaAI's P-Video-2-Pro-Replace is live on Runware on D0 🎬 Swap the people in a video you already have. Send a clip and one or two photos, and get the same video back with the new person: motion, camera and soundtrack kept, lips following the original audio. - Source clips from 1 to 120s, mp4, mov or webm, any aspect ratio from 1:4 to 4:1 - One photo replaces one person, two photos replace two, matched left to right or by prompt - 480p and 768p native, with 1080p, 2K and 4K upscaled from 768p - Pruna reports ~10s of processing per second of video at 480p, ~12s at 768p from $0.025 per second with 50% off until November 19 (regular from $0.05)

@GoogleDeepMind's Gemini Omni Flash 1.2 is live on Runware on D0 Google's update takes text-to-video to 15 seconds and edits source clips up to 15 seconds long. - Text-to-video from 3 to 15s, with native audio - Edit 3-15s source videos, with up to 7 reference images - Extend clips up to 30s long by 3-10s per call - Re-edit any output in a follow-up call using its videoId - 360p, 720p, 1080p and 4K, in 16:9 or 9:16 Already calling Gemini Omni Flash 1.1 on Runware? Switch your model ID to google:gemini@omni-flash-1.2 from $0.02 per second

We’re excited to welcome three engineering leaders from NVIDIA, OpenAI and NetApp to the Runware team; • Risto Haukioja, Head of Infrastructure Engineering. 8 years at NVIDIA, most recently on NIM • Matti Oikarinen, CISO. Frontier AI security at OpenAI, AWS Elastic File System architect at Amazon • Matt Baldwin, Principal Engineer, AI Infrastructure. Ran large-scale Kubernetes across NVIDIA DGX Cloud, founded StackPointCloud (acquired by NetApp) They’ll own orchestration, scheduling and security for the GPU fleet running on our Sonic Pods. Welcome aboard, Risto, Matti and Matt! Read the full announcement: runware.ai/blog/engineering-…

@OpenAI GPT-Image-2.6 Flare and Sunburst are live on Runware 🖼️ The 2.6 update adds animated GIF and SVG output to both models. Flare is for fast, interactive iteration. Sunburst is for detailed compositions and precise edits. - Text-to-GIF, text-to-vector (SVG) and text-to-image from the same model ID - Edit from marks drawn over an image, like a circle around the object to change - A thinking level sets how much the model plans before rendering, separate from quality - OpenAI reports Flare returns images in as little as 6 seconds at low quality Already calling GPT-Image-2.5 on Runware? Switch your model ID to openai:gpt-image@2.6-flare or openai:gpt-image@2.6-sunburst Sunburst from $0.0374 per image, Flare from $0.0587 per image

@Kimi_Moonshot's K3.1 Pro is live on Runware 🧠​ Moonshot AI’s post-trained successor to Kimi K3, built for coding agents, scripted 3D and CAD modeling, and financial research. - 1M-token context window, up to 1M output tokens - Always-on reasoning with configurable effort - Image input, tool calling and structured outputs - Moonshot reports 65.7% on Terminal-Bench 4.0 and 92.7% on BenchCAD with tools - OpenAI-compatible: call /v1/chat/completions with model moonshotai:kimi@k3.1-pro Already calling Kimi K3 on Runware? Same endpoint, just switch the model ID. from $2 per 1M input tokens and $10 per 1M output tokens

@Kling_ai Kling VIDEO 4.0 Flash is live on Runware 🎬 Kling Video 4.0’s variant for rapid 720p video generation and prompt-based editing, with outputs up to 20 seconds. - Inputs: text, a first frame, up to 10 reference images, up to 5 reference videos, up to 7 reusable elements - Edit an existing video by passing it as inputs.video with a prompt - 720p presets: 1280x720, 720x1280, 940x940, 1460x624 - Model ID: klingai:kling-video@4.0-flash from $0.094 per second

@krea_ai Krea 3 is live on Runware 🖼️ Krea's latest image model for reference-driven editing. Start from a product, a person, or a room, describe the change, and keep everything else. - Combine up to 8 reference images in one edit - Draw up to 24 boxes to set where an edit lands and which object to pull from each reference - 1K, 2K and 4K output across 10 aspect ratios, from 21:9 to 9:16 - Two thinking levels: fast by default, high for multi-step edits and multi-reference compositions from $0.025 per output megapixel

@topazlabs Wonder 4 is live on Runware 📷​ Topaz Labs delivers denoising, deblurring, compression artifact removal, face recovery and upscaling in their latest image enhancement model, Wonder 4. - Upscale 1x to 6x with upscaleFactor, or use 1 to clean up an image at its original size - Input and output up to 256 MP and 32,000 px per side - JPG, PNG or WEBP output Already calling Wonder 3.5 on Runware? Switch your model ID to topazlabs:wonder@4 from $0.0428 per image

@MiniMax_AI H3 Lite and H3 Highspeed are live on Runware 🎬 Two lower-cost tiers of H3 that take the same inputs. Try Lite for drafts and batch variations, Highspeed for everyday-use and social content. 4 to 15 second clips at 480p or 768p, six aspect ratios from 21:9 to 9:16 Text, a first frame, or a first and last frame pair Up to 9 reference images, 3 videos and 3 audio clips per request Already calling H3 on Runware? Switch your model ID to minimax:h3@lite or minimax:h3@highspeed from $0.025 per second on Lite and $0.04 per second on Highspeed. At 768p, that's $0.04/s and $0.06/s vs $0.08/s on H3

@typesafeai's Jev 1.14 is live on Runware 👁️ Jev's 1.14 update has the ability to read images and answer typed questions about them, plus a new Box question that returns where an object sits in the image. - Images go anywhere text goes: state, instructions, or criteria - Box answers: 4 edges as 0 to 1 fractions, plus confidence - PNG or JPEG, 280 to 1568px per side - Experimental: add the TypeSafe-Danger-Zone-Images: true header Already calling Jev on Runware? Switch your model ID to typesafe:jev@1.14 from $0.042 per 1M input tokens, images included, output free

@PrunaAI’s P-Video-Upscale is live on Runware! ​ Upscale an existing clip up to 4K with no prompt. Outputs retain the source aspect ratio and audio track. - Frame interpolation up to 60 FPS, set with fps (16 to 60) - Turbo mode for faster runs, at some cost to fine detail - Inputs up to 512 MB, output as MP4, WEBM or MOV from $0.005 per output second, with 50% off regular pricing until November 8

@ideogram_ai's 4.5 Precise Edit can now modify one region of a 4K or 8K image on Runware and return it at full resolution, with no quality degradation outside of the edited region! - New settings.editRegion: a box as [x1, y1, x2, y2] in seed image pixels, or "auto" with a maskImage - With a region set, the output keeps the seedImage's width and height - seedImage up to 67,108,864 px (8192x8192) - seedImage, referenceImages and maskImage up to 50 MB each, up from 25 MB from $0.008 per image

@ideogram_ai's Vector 1.0 is live on Runware 🖼️ Generate SVGs from a text prompt, or convert an existing image into one. Each shape stays editable, so you can match colors to a brand and refine individual elements in a vector editor. - Text to SVG, guided by up to 5 reference images via inputs.referenceImages - 3 quality levels and 15 2K presets, from 683x2048 to 2048x683 - Image to SVG from jpg, png or webp up to 20 MB - settings.mode for image inputs: simple for fewer shapes, detailed to keep fine detail - Runs as its own Runware taskType: vectorize from $0.10 per image

@PrunaAI’s P-Video-2-Pro-Avatar is live on Runware One portrait image plus an audio track or a text script returns a full lip-synced talking-head video. - 1 second to 3 minutes of output - Upload your own audio, or use built-in TTS with 30 voices across 10 languages - A text prompt steers motion and expression - Up to 4K, as MP4, WEBM or MOV from $0.01 per second, with 50% off regular pricing until November 5!

@OpenAI's GPT-6.1 Terra is live on Runware 🧠 A faster model for computer use, visual reasoning and high-volume work. - 1,050,000-token context, 128,000 max output - Text and image input, text output - Function calling, structured outputs, variable reasoning effort - Call openai:gpt@6.1-terra on /v1/chat/completions, add "stream": true for SSE Pricing from $1.00 per 1M input tokens and $5.00 per 1M output