Roadmap

Tools and studio surfaces we're building but haven't shipped yet. Anything on this list is either a prototype in the codebase, a verified gap in the workflow, or a scoped backend plan. We'd rather under-promise here than ship something that breaks mid-render.

Want one of these now? Hybrig is local-first — you can usually get the same result today by chaining local image generation with HyperFrames or your own ComfyUI graph.

  • Content-aware inpainting

    in development

    Mask + regenerate at full resolution. Type an instruction like "remove the watermark in the bottom-right" and Flux fills in plausible background. Backend wiring against fal.ai's Flux inpaint slug needs verification before this ships — early calls hit param-name mismatches.

  • Pose transfer

    in development

    Re-render your subject in the pose of a reference image. Lock the face, change the body posture. Backend depends on fal.ai's Flux Control Pose endpoint; param wiring needs an end-to-end pass before it goes live.

  • Image upscale (2x / 4x)

    in development

    Sharpen and 2x/4x upscale stills. Topaz is subscription-only now; Magnific revokes on cancel. The local-first version of this lives in the spokesperson pipeline already; the standalone tool needs the fal.ai clarity-upscaler param wiring confirmed before it ships.

  • Curate board auth repair

    in development

    The owner-only keeper/reject board exists, but logged-out requests currently redirect to /login even though the app does not ship a /login route. Fix the auth destination or hide the route before treating /curate as a dependable production page.

  • Storyboard and timeline workspace

    design

    A real production flow needs projects, shots, shot variants, selected keepers, dialogue, prompts, reference roles, and rough timing in one place. This is the missing middle between raw Assets and final Deliver.

  • Unified Generate workspace

    design

    Collapse scattered image, video, batch, Seedance, Wan, and pipeline mental models into intent-based modes: Still, Motion, Talking Clip, Training/Reference, and Batch. Provider choice becomes a control inside the mode, not another top-level page.

  • Hybrig Tested — editorial benchmark authority for local AI gens

    planned

    A public, receipts-first review hub for local image and video models. Stock-chart-style score timelines so you can watch a model's quality move over time, not just see today's number. Overlay model comparisons (Wan 2.1 vs 2.2 vs 2.3 plotted side by side across months). Wayback-style year filter to pull up scores as they were on any past date. Community-submitted scored renders with weighted-recency aggregation so old verdicts fade as new ones land. Public methodology page with every score linking back to the actual sample render — no black-box ratings. Editorial cadence: monthly model digests, quarterly head-to-heads, annual state-of-local-AI. The /stack page's "Why this stack, today" section is the v0 of this; the full timeline + comparison overlays + community submissions are the destination.