Agents work on their own always-on computer and leave the results as files.

AI that finishes the work, not just the answer

Claude Opus 5.5GPT ImageMiniMax H3 TurboOmniVoiceIdol (Image to 3D)

MiniMax H3-Image for precise edits, GPT Image 2.6, and FLUX.3 [klein]One prompt in, finished work out

Trusted by 758K+ users in 236 countries.

Made by the Giz Agent

One goal in, finished work out

Each result below is what the Agent delivered from the prompt shown: it chose the models, ran every step on its own computer and saved the files. Open the real results, or start from the same prompt.

Video

Product launch film

Storyboard, five scenes that keep the same shoe, voice-over and music, cut into one 30-second MP4.

Make a 30-second product launch video for a white leather sneaker with a green stripe called "Verde One". Use my product photo as the reference so the shoe looks the same in every scene. Five scenes: hero shot of the shoe, close-up of the stitching, a runner in a city at sunrise, the shoe on a skate ramp at sunset, end card with "Verde One - Walk light." Add an English voice-over and music, render one MP4 and save it to Drive.

MiniMax H3 Turbo
LTX-2.5
ffmpeg

Watch the film

Video

App demo and promo film

Real screen takes of giz.ai, b-roll, voice-over, music and captions, edited and checked into one 30-second MP4 from a single prompt.

Make a studio-quality 60-second launch film for the Giz Agent (giz.ai): "One prompt in. Finished work out." Aim for the best quality in every respect, like a launch video from a top SaaS company: story, pacing, motion, typography, voice, music, mix and captions. Real use, not mockups. Use your screen-capture skill so each screen take is directed: the camera zooms and pans to what matters, a smooth cursor moves and clicks, typing is visible, waits are shortened. Sign in first, off camera. In setup, collapse the left sidebar and close any popups so the account name and old chats stay out of frame. Record in English (locale en-US). Takes: 1. https://www.giz.ai/agents/agent%3agiz — type the goal "Make a 30-second launch film for a white sneaker" into the composer and send it, then show the Agent planning and working (tool steps appearing). Hold on the progress, shortened. 2. https://www.giz.ai/showcase/ — finished results: the launch film, the website, the game and the investor deck; zoom into each result. 3. https://neon-harbor-runner-jo475csn49.mtl3.giz.sh/ — play the game the Agent built. Brand: use the real giz.ai logo, never retype it as text: https://www.giz.ai/apps/static/giz/brand/giz-ai-logo-dark.png on dark backgrounds, https://www.giz.ai/apps/static/giz/brand/giz-ai-logo-light.png on light ones. Open on an animated logo reveal, keep a small logo watermark over product shots, and end on the logo with "One prompt. Finished work." and "giz.ai". Dark premium look, one accent color from the logo, one modern sans-serif font. Add 2 short cinematic b-roll shots with no readable text, a confident English voice-over, an upbeat music bed that runs to the last frame under the voice, and short captions in a lower band that never cover on-screen text or each other and appear exactly with the spoken line. Render one 1080p 16:9 MP4. Before you finish, check loudness, peaks, silence, every scene's frames and the captions, fix anything off, and save it to Drive.

Screen capture
MiniMax H3 Turbo
Voice-over
Music
ffmpeg

Watch the film

Images & edits

Rewrite a sign, change the season

Korean lettering in the same carved style on the same board, a snowy evening, and everything else left as it was.

In the attached photo, change the wooden sign text to Korean "모닝빈 카페" in the same carved brush-script style, and turn the scene into a snowy winter evening with warm light glowing from inside. Keep the building, bicycle and plants exactly the same.

MiniMax H3-Image

Open result

Images & edits

Same person, new outfit and place

Her face stays hers while the coat, the scarf and the Paris café at golden hour are new.

Keep the same woman from the attached photo with the exact same face; change the red raincoat to a navy wool coat with a cream knit scarf, and place her inside a Paris café at golden hour.

MiniMax H3-Image

Open result

Images & edits

Put a product into a scene

The person from one photo on a skate ramp at sunset, wearing the exact sneaker from a product shot.

Picture 1 is a person and Picture 2 is a sneaker. Show the woman from Picture 1, same face, riding a skate ramp at sunset and wearing the exact sneaker from Picture 2, green stripe included.

MiniMax H3-Image

Open result

Music & audio

A whole EP, start to finish

Concept and track list, full lyrics, four sung two-minute tracks with ACE-Step in one sound and key plan, cover art and liner notes.

Produce a 4-track synth-pop EP called "Neon Harbor" from scratch, saved in Drive folder Albums/Neon Harbor: write the album concept and a track list, write full lyrics (verse, chorus, bridge) for each track, generate every track with ACE-Step at about 2 minutes each, with vocals, in a consistent style, make square cover art with MiniMax H3-Image showing the title "NEON HARBOR", and write liner-notes.md with the concept, lyrics and credits.

ACE-Step
MiniMax H3-Image

01 Arrival 02 Neon Tide 03 Ferry Lights 04 Harbor Dawn Liner notes & lyrics

Music & audio

Two-host podcast episode

Script, two distinct natural voices, intro and outro stings, loudness-matched and mixed into one MP3.

Produce a 3-minute podcast episode called "Two Minds on AI" where two hosts (a calm female host and an energetic male host) discuss how small businesses can use AI agents. Write the script, voice each host with a different natural voice, add a short intro and outro music sting, and mix it into one MP3 saved to Drive.

OmniVoice
MiniMax Music
ffmpeg

Script

Websites & games

Launch website, live

Copy, lifestyle photos that keep the same shoe, a size guide and FAQ, designed and published on its own link.

Build and publish a polished one-page launch website for the sneaker "Verde One" (white leather, green stripe, tagline "Walk light."). Use the product photo Agents/Giz/2026-11-15T04-11-09-236Z-818f3709.webp and generate 2 more lifestyle images with MiniMax H3-Image that keep the same shoe. Sections: hero with a pre-order button, 3 feature cards, a lookbook gallery, size guide table, FAQ, footer. Responsive, modern, accessible. Publish it as a public site and reply with the public URL.

MiniMax H3-Image
Giz Apps
Browser check

Open live site

Websites & games

Playable browser game

An endless runner with keyboard and touch controls, sound, best score and generated title art, live on a public link.

Build and publish a polished browser game: "Neon Harbor Runner", an endless runner where a glowing boat dodges buoys in a neon harbor at night. Keyboard and touch controls, score and best score, increasing speed, sound effects, start and game-over screens, works on phones. Generate the title screen art with MiniMax H3-Image. Publish it as a public site and reply with the URL.

MiniMax H3-Image
Canvas + Web Audio
Giz Apps

Play the game

Docs & decks

Investor deck with real figures

Ten slides with researched market figures and sources, a chart and product images, as an editable PPTX and a PDF.

Create a 10-slide investor pitch deck for "Verde One", a sustainable leather sneaker startup: problem, solution, product, market size (research real public figures for the global sneaker market and cite sources), business model, go-to-market, competition, traction plan, team, the ask. Use a clean modern design with generated product images (MiniMax H3-Image) and at least one chart. Export it as PPTX and PDF to Drive and reply with both paths.

Web research
MiniMax H3-Image
Slides
LibreOffice

PDF PPTX

From a photo to a 3D model

One product photo becomes a model you can rotate, saved as a 3D file ready for a viewer or a scene.

Turn the sneaker in my product photo into a 3D model I can rotate (image to 3D), and save the 3D file to Drive.

Idol (Image to 3D)

Open result

What one Agent works with

Every AI modelOver 1,000 image, video, music, voice, 3D and language models; the Agent picks the right one for each step.Its own computerCode, a browser, video and office tools run on a cloud computer, so it can edit, render and check its own work.

Files you keepEvery result is saved in your Drive to open, edit, share, or hand back to the Agent for the next round.

Publish in one stepWebsites, apps and games go live on a public link, and any file can get a share link.

Runs on a scheduleTurn any task into an automation that runs daily or weekly and sends you the result.

Use it anywhereCall the same Agent and models from Claude, ChatGPT or your own tools over MCP.

Popular now

Input → output: text / image / video / audio → video / audio
Released: Jul 31, 2026
Capabilities: Text to video and stereo audio · First/end frame guidance · Person, prop and background references with a guide video · Control video · Injected keyframes · Soundtrack guidance · Control-video audio reuse · Audio generation from control video · Sliding-window continuation · Video extension · Optional AI 2× upscale · Optional faster step skipping
Output: Video + stereo audio
Resolution: 480p · 576p · 736p · optional 2× upscale
Duration: 4.5–20 sec

Fast MiniMax H3 video model for synchronized video and stereo-audio generation with first/end frames, control video, keyframe injection, soundtrack guidance, and long-window continuation.

Sperid LabsIdol (Image to 3D)

3DFree · 3 uses per dayAfter allowance · ≈ $0.055/use

Input → output: image → 3D
Released: Nov 11, 2026
Input: One object picture
Formats: PLY (Gaussian splat)

Turns a picture of an object into a 3D Gaussian splat you can turn and view from any side.

MiniMaxMiniMax H3-Image

ImageFree · 4 uses per 2 hoursAfter allowance · ≈ $0.022/image

Input → output: text / image → image
Released: Nov 13, 2026
Capabilities: Image editing · Up to 16 references · Up to 16 MP
Resolution: 1–16 MP
Formats: PNG · JPG · WEBP
References: Up to 16 images

High-resolution image generation and editing with up to 16 reference images.

k2-fsaOmniVoice

AudioFree · 4 uses per 2 hoursAfter allowance · ≈ $0.0055/use

Input → output: text / audio → audio
Released: Apr 1, 2026
Formats: WAV · MP3
Cloning: Zero-shot voice clone · Voice design

Massively multilingual zero-shot TTS with Voice Clone and Voice Design modes, prompt preprocessing, speed, duration, and diffusion controls.

PaddlePaddlePaddleOCR-VL 1.6

TranscriptionFree · 4 uses per 2 hoursAfter allowance · ≈ $0.011/use

Input → output: file / image → text
Released: May 29, 2026
Input: PDFs · Documents · Images
Output: Searchable PDF · Markdown

Create one searchable PDF from ordered slide images or an existing PDF, preserving the original design with invisible OCR text. Also extract structured Markdown with PaddleOCR-VL 1.6.

NVIDIANemotron 3.5 ASR Streaming 0.6B

TranscriptionFree · 4 uses per 2 hoursAfter allowance · ≈ $0.033/use

Input → output: audio → text
Released: Dec 24, 2024
Input: Audio / speech file
Output: Streaming transcription + timestamps

⚡️ Low-latency multilingual transcription | streaming ASR architecture | punctuation | capitalization | auto language detection

Video

MiniMaxMiniMax H3 Turbo

VideoFree · 2 uses per dayAfter allowance · ≈ $0.22–1.4/use

Input → output: text / image / video / audio → video / audio
Released: Jul 31, 2026
Capabilities: Text to video and stereo audio · First/end frame guidance · Person, prop and background references with a guide video · Control video · Injected keyframes · Soundtrack guidance · Control-video audio reuse · Audio generation from control video · Sliding-window continuation · Video extension · Optional AI 2× upscale · Optional faster step skipping
Output: Video + stereo audio
Resolution: 480p · 576p · 736p · optional 2× upscale
Duration: 4.5–20 sec

Fast MiniMax H3 video model for synchronized video and stereo-audio generation with first/end frames, control video, keyframe injection, soundtrack guidance, and long-window continuation.

LightricksLTX-2.5 Distilled

VideoFree · 2 uses per dayAfter allowance · ≈ $0.0413/sec (based on 4s)

Input → output: text / image / video / audio → video / audio
Released: Aug 11, 2026
Capabilities: Native multi-shot story · Two-stage quality · Automatic duration · Diffusion video decoder · Video extension · Ingredients · Inpainting · Outpainting · First/end frames · Exact keyframes · Pose control · Depth control · Canny control · Raw video control
References: First/end frames · keyframes · control video
Resolution: 480p · 540p · 720p · 1080p
Duration: 1–20 sec