Why Hybrig

ComfyUI is free and open source. Why not just use it?

You can. Plenty of AI tinkerers do. The engines under Hybrig — ComfyUI, Wan, Flux, PuLID, EchoMimic, F5-TTS, ffmpeg — are all open source and free. Hybrig is what you’d build on top if you wanted to ship client video instead of learn ML.

The shorter version: ComfyUI is to Hybrig what Postgres is to a finished SaaS app. Postgres is free, open source, and powerful. Nobody ships a finished product to clients on raw Postgres — they ship the application on top. Same here. The engine is free. The orchestration on top is the product.

Ten things the orchestration layer adds

Each item below is something the raw open-source stack can’t do out of the box. Each item is also something you could build yourself, given a season of weekends. The list is the value proposition stated honestly: this is what you’re paying for when you install Hybrig.

  1. 01 · The chain is pre-wired

    Flux + LoRA → Wan → F5-TTS → UGC Audio Fix → EchoMimic → HyperFrames. In a raw node editor that's a two-week learning curve plus a binder of YouTube tutorials. In Hybrig you click Run.

    See every layer →

  2. 02 · Per-prospect orchestration

    Drop your CRM export. Hybrig fans out 68 personalized renders from one row per prospect — name, zip, market merged into the script and the on-screen chrome. Open-source node editors render one video per click.

    See the render workers →

  3. 03 · HyperFrames variants with merge tags

    Branded intros, lower-thirds, outros baked locally with merge tags ({contractor_name}, {zip}, {market}). Save the template once, reuse on every prospect render. Raw open-source has no concept of templates with per-prospect variables.

    Compose at /hyperframes →

  4. 04 · A pluggable LLM brain

    Local Ollama by default for script generation, prompt expansion, and CadenceMeter cadence checks. Cloud Anthropic or OpenAI when you ask for them — never the default. Raw open-source ships zero LLM orchestration.

    Pick your brain →

  5. 05 · Real product UX

    /studio for the node canvas, /farm for the prospect queue, /pipeline for chained runs, /create for the one-click spokesperson tool, /dashboard for render history, /settings for routing. Raw open-source ships only the node editor — every other surface is yours to build.

  6. 06 · Cloud rescue valve, not the default

    When local misses on a critical shot, Hybrig auto-falls through Wan 2.7 → Hunyuan → LTX → cloud Seedance. Veo, Runway, Hedra, and Anthropic are wired the same way. Open-source local-only setups have no rescue path; if Wan misbehaves at 2 a.m., the queue drops.

  7. 07 · Saved Variants, character packs, voice profiles

    Every face, every voice, every wardrobe is a first-class object you save once and reuse forever. Raw open-source asks you to rebuild the conditioning chain on every render.

  8. 08 · Doctrines baked into the chain

    CadenceMeter prevents robotic-prosody voice output. The StoryBrand doctrine guides scene structure. The Billie Jean creative bar nudges scripts past the standard AI-spokesperson tropes. These are system-prompt layers built into the orchestration — they don't ship in any node editor.

  9. 09 · Per-prospect data injection

    Render 68 videos that each say the contractor's company name, zip, and offer code in the script and on the lower-third. Same engine underneath; the orchestrator merges the variables. Without this layer you'd be writing 68 scripts by hand.

  10. 10 · A different brand and audience

    Hybrig is for creators making personalized client video at scale — agencies, content farms, founder-marketers, indie filmmakers shipping outreach. Open-source node editors target AI hobbyists who want to learn ML. Same engine, different product on top. Different audience, different brand.

One prospect video, both ways

A contractor wants a personalized 30-second pitch video sent to 68 prospects on his lead list. Same engines underneath. Different product on top.

StageRaw open-sourceHybrig
Open the projectBoot the node editor. Reload last weekend's graph. Hope no node updated and broke the wiring.Click Hybrig. The chain is pre-wired. Last week's settings are still there.
Drop in this prospectManually copy the contractor's name, zip, and offer code into the script node. Repeat for the lower-third overlay node.Drop one row of CSV. Merge tags fill themselves in: {contractor_name}, {zip}, {market}.
Render the stillWire Flux + LoRA + PuLID conditioning. Watch out: if you have both PuLID and Redux nodes attached, Flux will conflict on the conditioning slot.Hybrig picks the right identity layer for the model and silently disables the conflicting one.
AnimateSwitch to the Wan 2.7 graph. Re-feed the keeper still. If Wan 2.7 misbehaves on this install, manually rewire to an older model.Auto-fallback chain runs Wan 2.7 → Hunyuan → LTX → cloud Seedance. The queue doesn't drop.
VoiceSwitch to your TTS tool. Run the script. Hand-tune cadence to match a 6-second clip. Repeat if the voice sounds robotic.F5-TTS clones your voice locally, CadenceMeter checks prosody, UGC Audio Fix smooths it. All on rig, no upload.
LipsyncOpen EchoMimicV2 graph. Probe whether the right .pth files are in models/echo_mimic/v2/. If not, switch to LatentSync. If that's broken too, give up and use cloud sync-lipsync.Pipeline probes /object_info, picks the working local lipsync, falls back to cloud only if both local options are missing.
Brand chromeOpen ffmpeg manually. Hand-write the overlay filter chain for the logo intro and lower-third. Hope your font path is right.HyperFrames bakes the branded intro, lower-third, and outro automatically. The merge tags from step 2 land in the lower-third.
Now do it 67 more timesOpen the next CSV row. Repeat every step above. By prospect 14 you're tired and start cutting corners.The prospect queue already ran 67 more renders in parallel across your fleet. Folder of finals is on disk when you wake up.

The bottom line

Open-source AI tools render one video per click. Hybrig renders one per prospect. The engine is the same. The orchestration on top is the product.

DIY-impossible at scale

You can absolutely wire the open-source stack yourself for one video, one weekend, one prospect. Plenty of creators do, and they ship great work that way. The scale is the cliff. At one video per click, the node editor is fine. At 68 videos per prospect list, per week, with merge-tagged scripts and per-prospect lower-thirds, the manual rewiring becomes the job. The job stops being client work and starts being graph maintenance.

Hybrig is the orchestration layer that lets you keep the engines and lose the maintenance. Local by default — the GPU you already own renders every job. Cloud only as a rescue valve when local misses on a critical shot. Per-prospect fan-out, branded chrome, doctrine-checked voice, all baked in. Same models the open-source community ships, finished into a product.

The other half of the math, since this is the page that earns the honest comparison: cloud rivals like Enhancor ship the same engines on rented GPUs and bill $50–$200/month for the privilege. Hybrig ships them on the GPU under your desk — free per render after install. Same engine, different distribution model. That part is structural, not a feature.

Try the assembled product

The desktop app installs in one click. It bundles the engines, points the wizard at your GPU, and signs you in to your Hybrig account. The first render is on the rig you already own.

For solo creators

One person, one rig. Read the creator pitch →

For agencies / farms

68 prospects per week. Read the farm pitch →