A little guidance. More room to create.

New Google accounts receive 60 welcome credits, valid for 30 days from the grant. No payment card is needed to use these credits. Text-to-video costs 6 credits per task; image-to-video and first-and-last-frame cost 8. Sign in to check your actual balance and expiry.

The 60-credit welcome grant is a one-time new-account benefit, not a daily allowance. With no other charges, 60 credits cover up to 10 text-to-video tasks or 7 image/first-and-last-frame tasks with 4 credits left. Mixed usage reduces the same balance. Unused welcome credits expire after 30 days.

The current H3 workflow and download flow do not add a brand watermark. The model may still generate text or marks in the image; completely mark-free results are not guaranteed.

Welcome credits expire 30 days after they are granted, rather than at the end of the calendar month. Check Account & credits for the grant and expiry records. The welcome grant is not a subscription and does not automatically refill.

Use image-to-video with a clear, unobstructed portrait. Start with subtle movements such as blinking or smiling, short clips and a fixed camera. Large turns and occlusions may change the face. There is no cross-task character lock.

Describe a specific action and its pace instead of repeating the appearance: for example, a gentle blink and smile with a fixed camera. Separate subject movement from camera movement. Identical start and end images can constrain motion.

Complex actions, fast motion, occlusions and large viewpoint changes can cause artifacts. Try simpler actions, shorter clips and clearer references. Prompts cannot guarantee distortion-free output.

Use a clean product image with a clear silhouette and room around it. Try keeping the product still while the camera slowly pushes in or light sweeps across it. The site generates footage, not a finished ad with script, captions and voice-over.

Reduce rotation, occlusion and large movements. H3 does not provide pixel-perfect product or text locking. Add critical text in editing and check each frame. Generated footage is not proof of real product performance.

Specify a stationary subject and one camera move, such as a slow push-in without cuts. Test pans, orbits and push-ins separately. Camera movement is prompted, not controlled by a precise path editor.

Both English and Chinese can be entered; results require testing. For text-to-video, describe subject, action, setting, camera and light. For image-to-video, focus on motion. The current limit is 2,000 characters.

The public tool does not accept a reference video or extract its camera path. You can describe the motion and use first and last images to guide the opening and ending composition. This is visual guidance, not motion tracking; it does not guarantee the same timing or trajectory as the reference.

Describe the starting shot, the forward move and the final framing: “Begin in a medium two-shot. The camera slowly dollies forward into a close-up of the face and dragon snout, one continuous shot, no orbit.” If you have both compositions, use first-and-last-frame mode. Avoid adding a pan, orbit and zoom-out to the same brief.

Select first-and-last-frame mode, upload the start and end images and describe the transition. PNG, JPEG and WebP are supported, up to 10 MB each. Similar subjects and framing are easier starting points. Frames are processed by the model, not replayed pixel for pixel.

Large differences between inputs or conflicting viewpoints can cause warping. Try the same subject in similar compositions and one continuous action, or split the sequence. Exact intermediate and final frames are not guaranteed.

The current single-task options are 5, 10 and 15 seconds. H3 rounds frame counts, so file duration can vary slightly. Generate separate shots and edit them together. Reusing the last frame can help continuity but does not guarantee a seamless join. One-click extension is not available.

Select First & last frame, upload the opening image to the first input and the ending image to the second. Describe the path between them, such as a steady camera push from a medium two-shot into a close-up. Choose the duration and orientation. Both images are required; swapping their order changes the intended direction.

The public generator currently saves silent MP4 clips. Audio generation, voice-over and lip-sync controls are not connected in this workflow. Some homepage showcase clips were created with a separate audio-capable workflow; their sound is not a promise that every generated clip includes audio. Add your own licensed sound after downloading.

The public generator offers landscape (896×512) and portrait (512×896) MP4 at 24 fps. The 16:9 and 9:16 labels describe approximate orientations, rather than exact pixel ratios. There is no public 1080p, 1440p or 4K output switch. Selected homepage samples were separately upscaled to 2560×1440.

Check the destination platform rules, AI disclosure requirements, model license and rights to uploaded material. The site does not provide a blanket commercial license or guarantee publishing, monetization or ad approval.

When the task succeeds, preview the result and use its download button to save the MP4. Keep a local copy of work you want to retain. If downloading fails, confirm you are signed in to the account that created the task and retry from its result or history page.

The fantasy homepage video was separately generated and upscaled with a dedicated workflow. The public generator currently returns its standard-resolution output; there is no public upscale switch. The showcase demonstrates a selected result, not a guaranteed resolution or level of detail for every task.

Speed depends on GPU load, queue length, duration and content. There is no guaranteed completion time. Check the studio or task history before submitting again; a long wait does not itself mean failure.

Current costs are 6 credits for text-to-video and 8 for image-to-video or first-and-last-frame. Refund logic runs when a task is explicitly recorded as failed with a linked credit charge. Closing the page, a timeout or disliking a successful result does not automatically trigger a refund.

There is no automatic free retry for an unsatisfactory successful result. A new submission costs credits again. Start with short clips and simple actions, and review inputs before submitting.

The current charge is per task and mode: 6 credits for text-to-video, 8 for image-to-video or first-and-last-frame. Selecting 10 or 15 seconds does not currently multiply that charge. Longer clips can take more GPU time and may be harder to keep visually consistent. Check the displayed cost before submitting.

A submitted task runs on the remote generation service, not in your browser. Closing the page does not cancel it. Sign in again and open the generator to recover an active task, or check Library & history. Browser polling updates the displayed status; a closed tab may delay status updates and failure reconciliation until you return.

Each account can have one active ComfyUI video task at a time. If a task is still pending or processing, the site returns that existing task instead of creating another one. Opening more tabs does not add extra concurrent slots. Wait for a confirmed final status before submitting the next task.

A network interruption or missing upstream history can prevent the site from confirming whether a task was accepted or finished. This does not prove failure. Do not repeatedly resubmit; check history and allow the next status check to run. Keep the task ID if you need help diagnosing a persistent issue.

A submitted workflow keeps the prompt and inputs it was created with. The current public interface does not offer a cancel button or editing of a running task. Changing the form is not a cancellation. For a different result, wait for the current task to finish and then submit a new task with the revised inputs.

Only upload material you have permission to use, with any necessary consent. Inputs are sent to the connected ComfyUI service. There is no uniform automatic deletion period promised here. Avoid private or confidential material unsuitable for that service.

Generated video delivery checks the signed-in user and task ownership. A result URL is not a public sharing page. Download the MP4 and share the file through your preferred channel if you have the necessary rights. Public showcase videos are separate from private task results.

Choose text-to-video when you have a written scene but no starting image. Choose image-to-video when a photo should define the subject and composition. Choose first-and-last-frame when you have both opening and ending compositions. All three modes use the same account balance.

Use Sign In and choose Google. FreeVideo AI currently supports Google sign-in only; there is no email/password registration or password-reset flow. Use the same Google account to find your balance and tasks on another device.

Sign in, select a mode, and write a short prompt describing one scene and one action. Add reference images if your mode requires them. Choose 5, 10 or 15 seconds and landscape or portrait. Check your balance, submit once, then wait for the preview. Start with 5 seconds to evaluate your direction.

Try: “A silver spacecraft glides past a vast ring-shaped orbital station above a blue planet. Wide establishing shot, slow forward dolly, crisp daylight, realistic cinematic materials, one continuous shot.” Name a concrete subject, one action, a setting and one camera move. Specific visual details are more useful than repeating “high quality.”

No image is required in text-to-video mode. Describe the subject, setting and motion in the prompt. If the result needs to match a particular face, product or composition, switch to image-to-video and supply a suitable reference instead.

For a short clip, begin with one scene and one continuous action. Several locations, costume changes or conflicting camera instructions can make the result unpredictable. For a longer story, generate individual shots and assemble them in an editor. There is no storyboard timeline or multi-shot editor in the current tool.

The current interface has one prompt field and no user-facing negative prompt or seed control. You can briefly state unwanted behavior in the main prompt, such as “no cuts, no text,” but this is not a guarantee. Reusing the same prompt does not guarantee an identical result.

Choose Image to Video, upload one PNG, JPEG or WebP of 10 MB or less, and describe what should move. For example: “The woman gently blinks and gives a small smile. Her head stays steady; the camera is fixed.” Select a duration and orientation, then generate.

Use a sharp image with a clear main subject, visible important details and some room around the subject. Avoid heavy blur, tiny faces, covered features and busy backgrounds when those details must stay consistent. A higher-resolution upload cannot guarantee additional output detail; the workflow resizes the reference for generation.

Matching the reference to the selected landscape or portrait orientation helps you plan the framing. Leave space around faces, hands and products so important details are not close to the edges. The model processes the image at its working resolution, so borders and framing may differ in the result.

You can request it in the prompt, for example: “The person remains still while rain falls behind them; fixed camera.” Keep the requested change simple and review the result. There is no motion brush, region mask or pixel-level subject freeze in the current interface.

Check that the file is PNG, JPEG or WebP and no larger than 10 MB. Renaming a HEIC or other file to .jpg does not convert it; export a real supported image first. In first-and-last-frame mode, add both images. If the file meets these requirements, check your connection and task status before resubmitting.