MiniMax Voice Cloning clones a voice from a sample audio clip and returns a custom voice ID for use in speech synthesis, with optional noise reduction, volume normalization, and a text preview of the cloned voice. Billed per cloning request.
MiniMax H3 Max Reframe Video expands existing videos to a selected aspect ratio with generative outpainting, preserving source audio and timing. Longer source videos are processed in overlapping windows, and billed duration can exceed the final video duration. Billed per second of generated video.
MiniMax H3 Max Relight relights any video from a lighting sphere image, while preserving the source subjects, motion, camera, and audio. Billed duration matches the source video length, at 480P to 2K resolution. Billed per second of generated video.
MiniMax H3 Max Recast recasts the people in a video using reference photos while preserving the source motion, camera, cuts, and audio. Supports optional realism enhancement at 768P or 1080P. Reference images are included at no additional charge. Billed per second of output video.
MiniMax H3 Max Insert Video inserts a new scene (5 to 13 seconds) into an existing video, guided by a prompt, reference images, or reference videos, with optional color matching, then returns to the original footage. Only the newly generated scene duration is billed, not the full output video.
Kwaivgi Kling Create Voice creates a custom voice ID from a 5 to 30 second clean single-voice audio clip, for use with Kling voice control in video generation. Supports mp3/wav audio or mp4/mov video input. Billed per voice creation request.
Topaz Upscale Image Precision is Topaz Labs' professional photo upscaling model powered by Gigapixel precision models (Standard V2, High Fidelity V3/V2, Low Resolution V2, CGI, Text Refine, Faces). It enlarges images faithfully up to 4x with optional face enhancement, sharpening, denoising, and compression artifact removal.
Kling 4.0 Flash Video Edit is Kwaivgi's video-editing model: give it an input video (@Video1) and prompt instructions describing the edits shot by shot (up to 8 shots with start and end timestamps), and it regenerates the footage to follow the new script while keeping the narrative structure of the original. 720p output; output duration defaults to the source clip's length; audio is generated natively.
Kling 4.0 Flash Text to Video is Kwaivgi's fast text-to-video flagship: describe any idea and it produces coherent 720p footage with sharp prompt following and fluent physics. Duration 3-20 seconds; aspect ratio auto/16:9/9:16/1:1/21:9 (auto adapts to the prompt); audio is generated natively.
Kling 4.0 Flash Reference to Video is Kwaivgi's multi-reference video model: generate a coherent 720p video from a prompt plus multi-reference assets, locking subject identity across shots. Up to 10 reference images, up to 5 reference videos (each 3-20s, combined with element videos within a 20s budget), and up to 7 elements (a frontal image, 1-3 multi-angle references, or a 3-8s element video); voices not accepted. Aspect ratio auto/16:9/9:16/1:1/21:9; audio generated natively.
Kling 4.0 Flash Image to Video is Kwaivgi's image-to-video model that takes a single image as the first frame and animates it: the prompt describes the shot (Flash does not take an end frame). Duration 3-20 seconds, 720p only, native audio generation.
GPT Image 2.6 Sunburst Text to Image is OpenAI's precision-focused image model, built for premium visual work with extra fidelity on intricate details, in exchange for longer generation times. Quality has four tiers (auto/low/medium/high) with a separate reasoning_effort control; output supports JPEG/PNG/WebP/GIF and SVG; up to 10 images per call.
GPT Image 2.6 Sunburst Image to Image is OpenAI's editing model built for the tightest control: edits scoped precisely to the instruction, with subject and composition preserved across many rounds of revision. Up to 16 reference images; mask or markup (annotated overlay) local editing; a frame pipeline renders stills to a single GIF loop (SVG unavailable here). Quality has four tiers plus a separate reasoning_effort control.
GPT Image 2.6 Flare Text to Image is OpenAI's image model for most applications: fast, high-quality generation with natural lighting, rich textures, and support for complex layouts including transparent backgrounds. Quality has four tiers (auto/low/medium/high) with a separate reasoning_effort control; output supports JPEG/PNG/WebP/GIF and SVG; up to 10 images per call.
GPT Image 2.6 Flare Image to Image is OpenAI's fast, high-quality editing model: it changes only what is asked, keeping subject, composition, and background intact, with reference subjects staying recognizable across successive edits. Up to 16 reference images; mask or markup (annotated overlay) local editing; a frame pipeline renders stills to a single GIF loop (SVG unavailable here). Quality has four tiers plus a separate reasoning_effort control.

BizyAirPlus
Run ComfyUI on Cloud GPUS
No local GPU needed. Install the plugin and run any workflow in the cloud.
Get the PluginAfter installation, start ComfyUI and choose BizyAirPlus mode from the run button to begin.