Compare the new image, video, voice, music, and transcription routes by their actual inputs and billing units; treat each node’s live quote as authoritative because references, duration, output size, and provider pricing can change the total.
Where: Flowgen → select the matching image, video, voice, music, or Audio Transcription node → model/operation controls and live quote.
Before you start
- Use the operation-specific picker. An image model, video model, TTS model, transcription model, and music operation are not interchangeable.
Key ideas
- Qwen Image 3 Pro is distinct from Qwen Image 3: The image picker offers Qwen Image 3 Pro as its own Kie route alongside Qwen Image 3. Pro exposes 1K and 2K, one generated image, PNG/JPEG, a seed, and prompt enhancement. It can take up to three connected reference images; a reference selects the edit route and changes the quote. The base quote is 4 Flowies at 1K or 7 at 2K before reference adjustments.
- FLUX 3 Max is a high-quality video route: FLUX 3 Max supports 5–20-second video at 720p or 1080p and can use text, start/end frames, or ordered keyframes. Generate audio is optional. Its list-basis quote is 18 Flowies per requested second at 720p and 31 at 1080p; current supplier promotions do not redefine the product quote.
- MiniMax H3 Max and H3 Max Turbo are separate tiers: Both support text or strict start/end-frame generation, including an end-frame-only request, plus an optional pinned soundtrack. Requested duration is 1–15 seconds and the rendered file can be about 0.7 seconds longer than requested. Per requested second, H3 Max costs 6/9/17 Flowies at 480P/768P/1080P; H3 Max Turbo costs 3/5/9. The visible quote uses stable supplier list rates, not temporary discounts.
- Seedance 2.5 normalizes source-bound output settings: When a Seedance 2.5 request uses strict first/last frames or edits/recreates a source video, fixed aspect selections from older saved canvases are normalized to Auto so output inherits source geometry. Edit/recreate requests also send the provider’s automatic duration setting. Keep user-facing duration numeric for quote calculation; the provider token is not a price value.
- Vidu Q4 Preview starts from an image: Vidu Q4 is not text-to-video: connect a source image or reference images. It supports 3–16 seconds, 540p through 4K, generated sound, up to 15 reference images, and up to three audio references (each 3–12 seconds). Its video quote is 4/8/10/16/33 Flowies per output second at 540p/720p/1080p/2K/4K.
- Azure MAI Voice and transcription use different units: MAI Voice 2.1 and MAI Voice 2.1 Flash use Azure-managed voices, supported speaking styles, and MP3/WAV output; choose voice language, voice, style, and format. They cost 1 Flowie per started 250 text characters, up to 5,000 characters. MAI Transcribe 2 is a separate uploaded-file node: WAV/MP3/FLAC up to ten minutes, Sweden Central, 1 Flowie per started measured audio minute. It can emit a text transcript and separate timing/speaker JSON.
- ElevenLabs audio prices are operation-unit based: The fal-backed ElevenLabs v4 and v4 Turbo voices cost 3 and 2 Flowies per started 250-character block (audio tags count; Turbo has a 5,000-character limit). SFX v2 costs 1 per started four output seconds (0.5–22 seconds). Music v2.5 costs 67 per started output minute: 3–60 seconds is one unit and 61 seconds starts a second. Auto duration reserves the maximum and settles measured output within the displayed quote.
- Suno V6 offers three model variants and operation-specific controls: The current model choices are V6, V6 Mini, and V6 Wild; older saved choices remain identifiable as legacy rather than new picker options. Available controls depend on operation: generation can use custom lyrics/title/style, instrumental, eligible duration, advanced negative tags, vocal gender, persona, variety, and style/weirdness/audio weights. Sounds expose seamless loop, BPM, key, and lyric capture. Extend and replace-section controls are tied to a source track.
- Quotes are live, not a permanent price list: The number shown before a run is the current product estimate. Model route, resolution, duration, connected references, generated outputs, operation, and measured media can change it. Review the exact node quote immediately before spending; an old article, screenshot, or provider promotion is not a substitute.
Steps
- Choose the job and add the matching node: Universal Image, Video Generation, Voice, Music/SFX, or Audio Transcription.
- Pick the exact model or operation. Inspect the input handles and any disabled controls before connecting media; for example, Vidu requires images while H3 Max accepts its strict start/end-frame pattern.
- Set model-specific options such as output size, duration, resolution, style, format, references, or audio. Do not assume a control carries over when you switch models.
- Read the updated quote on the node after all options and references are set. Check whether its unit is per image, requested video second, started character block, started output interval, or transcription minute.
- Run one representative output and review it before increasing resolution, length, or batch count. Use a new quote after changing inputs or model.
- For transcription, confirm the uploaded audio duration has finished loading; if the server reports a changed measured-minute quote, review the refreshed amount and deliberately retry rather than assuming the original charge.
Current media quote reference
- Qwen Image 3 Pro: 4 Flowies/image at 1K or 7 at 2K as the base; up to three references alter the total.
- FLUX 3 Max: 18 Flowies/requested second at 720p; 31 at 1080p.
- MiniMax H3 Max: 6/9/17 Flowies per requested second at 480P/768P/1080P. H3 Max Turbo: 3/5/9.
- Vidu Q4 Preview: 4/8/10/16/33 Flowies per output second at 540p/720p/1080p/2K/4K.
- MAI Voice 2.1 and Flash: 1 Flowie per started 250 text characters. MAI Transcribe 2: 1 per started measured audio minute.
- ElevenLabs v4/Turbo: 3/2 Flowies per started 250 characters; SFX v2: 1 per started four output seconds; Music v2.5: 67 per started output minute.
- Suno Studio has a separate quote for each operation. The live node quote is authoritative; do not multiply or infer another operation’s row.
Audio transcription output settings
- Timing: word, segment, or none; none does not fabricate timestamp values.
- Transcript style: verbatim or clean.
- Optional speaker labels, language hint from the supported language list, and one phrase hint per line.
- The text_out handle carries the transcript; json_out carries timing/speaker data when produced.
Suno V6 control map
- Generate: Simple or Custom mode, instrumental, title/style/lyrics, optional eligible duration, and expandable performance controls.
- Performance controls: negative tags; vocal gender where applicable; persona ID and style/voice type; variety; style weight; weirdness; audio weight.
- Sounds: seamless-loop toggle, tempo (20–300 BPM), key, and capture lyrics.
- Extend: source task/track and continuation time; Customize continuation reveals prompt/title/style choices.
- Replace Section: start/end plus full lyrics; the replacement span must be 10–60 seconds.
- Other operation groups include Create, Transform, Extract, and Publish. Their required source audio, task ID, or track ID changes by operation.
Seedance 2.5 source-bound requests
- Start/last-frame and source-video edit/recreate requests inherit source aspect instead of forcing a conflicting saved fixed ratio.
- Edit/recreate requests use the provider’s automatic duration while retaining numeric user-facing duration for quote calculation.
- Reference-image requests that are not strict frames and text-to-video can retain their selected fixed ratio.
Tips
- Keep generated voice identities with their originating provider/model: Azure managed voices, fal stock voices, and a user’s private ElevenLabs profile are different identity namespaces.
- For a fair model comparison, hold source image/audio, requested duration, resolution, and reference count constant, then compare the live quotes.
Limitations and important notes
- Model availability, supported inputs, price units, and provider availability can change; the live picker and quote take precedence.
- MAI Transcribe 2 file transcription is qualified in Sweden Central; do not assume another region or streaming transcription uses the same contract.
- A quote does not guarantee provider success or subjective quality.
Troubleshooting
The amount changes after connecting a reference or selecting a larger output.
That route prices more than its base generation. Re-read the current quote after all references, resolution, duration, and output count are set.
A model does not accept a text-only request.
Check that model’s required frame/image/audio handles. Vidu Q4 requires image input; selecting it does not imply text-to-video support.
A voice ID is rejected after changing providers.
Choose a voice from the matching provider/model catalog. Do not send a private ElevenLabs ID to fal or an Azure managed voice to another route.