Find your next starting point.

Casts, looks, voices and whole production methods to build your next piece on.

Browse without an account. Sign in to create, acquire, save or participate. 18+ resources require an adult account with mature content enabled.

• Exact-size IAB display and social ad images

  All 16 retained ad canvases, eight IAB display sizes, twelve platform packs; explicit model generation geometry; ordinary Comfy finishing graphs; native exact-pixel PNG/JPEG output and provenance.

• Retained ad recipes: UGC, interview, unboxing, restyle and product-spotlight treatments

  25 recipe treatments (persona, set, shot list, duration, CTA) as ordinary Comfy stages: Z-Image character references, Qwen-Image-Edit 2511 scene plates from the supplied product photo, ElevenLabs exact speech with fit checks, native H3 Ref2VA takes with anchored on-camera speech, optional ElevenLabs music and a native composition timeline built from measured outputs.

• Source-preserving analog video and decorative print screening

  Two native CPU methods with explicit source-pixel protection, seeded analog chroma/scanlines/grain, decorative RGB CMY dot screening, preserved sources, targeted parameter revision and separately exported reuse.

• Authored motion graphics, captions and finishing

  Build frame-based motion graphics, exact typography and captions around owned media and authored audio.

• Creative packs: carousels, DCO matrices, hook variants, multi-aspect masters, photo motion and shorts

  Twelve retained creative-pack originals as explicit finite items: H3 image-to-video from supplied photos, exact ElevenLabs speech, product-campaign still layouts, measured composition finishing, CPU source cuts and a fail-closed package of every item with its editable sources.

• Directed audio: songs, jingles, voiceover, podcasts, audiobooks and spots

  Six executable native Comfy ports with explicit caller-authored scripts, voice and music sources, native ducking/mastering, duration-checked spots and saved audio. Original provider/model intent retained.

• Directed still ads, ad sets, concept batteries, meme cards and thumbnails

  Six retained directed-still workflows as one ordinary Comfy graph per cell: product-grounded Gemini 3 Pro Image, literal TextOverlay copy, supplied logo and exact destination pixels and byte caps; independent ad-set hypothesis cells and concept lanes.

• Directed video: narrated, source-grounded, remix, lyric, film and desk productions

  16 general directed-video originals as staged native programs: exact TTS narration with speech-fit gates, cited source quotes, measured reference-ad structure, reviewed plates, self-hosted H3 I2V/Ref2VA shots, CPU edits of supplied video and native composition finishing with measured captions, cards and audio roles.

• Photographic ads and 12 image-to-video motion styles

  Animated banners default to video-first production: reviewed keyframe, H3 clip, then HTML5 image atlases with editable copy. Explicit static and element/photo treatments remain supported. 36 distinctive editable styles: 72 unchanged-photo demonstrations across eight product families plus 20 reference-conditioned Gemini 3 Pro Image variants across 12 styles and seven selected product scenes. Original product photos, model prompts/receipts and rejected revisions are preserved. Google HTML5/still, Facebook MP4/still and owned-web delivery use the same native destination checks. The selected I2V method adds twelve photographic directions across six products and 36 destination exports, with measured JPEG-atlas budgets and complete finite/reduced-motion endings.

• Feature film: one request to a finished multi-minute, series-grade film

  A native program that writes, casts, plates, voices, directs, keyframes, animates (self-hosted H3), reviews, edits, sound-designs and finishes a multi-minute film from one sentence. Every generation is an exported Comfy graph run on owned hosted compute runtimes; every creative choice is a model decision on actual text, pixels or sound, with receipts and resume.

• H3 4K video: text, image, first/last frames and references

  Explicit native H3 graph revisions generate on the required 32-pixel grid and crop to exact 3840x2160 delivery while preserving original model and media paths.

• H3 first/last-frame chains, seam assembly and cartoon film restyle

  Two retained H3 first/last-frame chain originals (parallel and sequential) and the cartoon film door, as per-keyframe stage graphs (Gemini partner stage or a self-hosted ladder), restage graphs, 192-frame H3 window graphs, the FILM seam graph and native handoff, seam and cartoon window tools.

• H3 long-take continuous shot and audiovisual continuation graphs

  Long continuous shots, including 60-second plans: bounded H3 windows with direct or saved motion-context AV latents, encoded AV/frame guides, last-frame hybrid continuation and exact overlap assembly. Includes an inventory of all 12 retained extension/long-take catalog originals.

• Self-hosted H3 finite image-to-video

  Selected H3 Ref2VA finite clips and source-built engine I2V canvas/native-original preservation checks.

• Purpose-based images, exact copy and structural editing

  Purpose-based photographic, exact-copy, source-mark and diagram methods from a bounded retained comparison; Z-Image Turbo native text-to-image, explicit upscale/sharpen topology edits and selected OpenAI gpt-image-1-mini generation.

• Music video: song-first singing performance or soundtrack montage

  Editable supplied-song and original-song-first Comfy graphs: global song-time slices condition native H3 Ref2VA with recurring performer references; exact frame assembly uses the original master. Explicit montage mode preserves soundtrack-led legacy intent without claiming singing.

• Native format cards: PNG stills and supplied product images

  Eight legacy card formats rendered from literal authored copy and optional supplied product imagery through native CLI and shared media.compose PNG delivery.

• Pre-production: identity plates, grounded references, angles and views, scripts and storyboards

  Optional script critique, rewrite, storyboard and identity-plate techniques. creative.develop is an optional structured development helper; reference-led planning can help keep the same character across scenes.

• Saved-voice H3 presenter ad with product references, captions, CTA and targeted revision

  Per-shot exact lines in a saved voice, measured fit to H3 17k+5 takes, Ref2VA takes with presenter + product references and anchored speech, sync sheets, composition finishing and a one-shot revision that reuses unchanged takes.

• Photographic product composition and campaign variants

  Declared crop/scale/composite of unchanged product pixels, authored EN/FR copy, carousel groups, ZIP export and targeted revision reuse.

• Original Qwen-Image FP8 text-to-image and controlled revisions

  Exact original non-GGUF Qwen-Image transformer, encoder and VAE: four current H100 outputs covering a product still, exact-word poster, prompt revision and independently exported project reuse.

• Verified synthetic voice assets and H3 speech guidance

  Synthetic 17.6/87.957-second sample clones, explicit provider-verification receipt, exact new TTS, independent asset import, and an H3 guided presenter (its example take is 124 frames; take length is a choice).

• Text-to-SFX packs: ElevenLabs by default, LTX-2.3 audio-only by choice, with measured gates

  Explicit per-effect briefs on ElevenLabs sound effects (default; native WavemakerElevenLabsSfx or hosted elevenlabs.sound_effects) or self-hosted LTX-2.3 audio-only (route='ltx-audio'), exact-duration trims and a native check for duration, non-silence, clipping and distinctness.

• Authored speech, original music and exact delivery

  Speech/music authoring and performance: sing exact lyrics with synchronized mouth and changing face through inspected graphs, provider routes and composition. Listen to the actual final soundtrack with owned audio, media.analyze measurements and an available audio-perception review route.

• SDXL keyframes, Ideogram typography posters and coherent brand icon sets

  Three original model routes with retained PNG originals, six explicit icon outputs, and native exact-size/file-cap finishing.

• Original subject LoRA training and reuse

  Original robot dataset, selected Z-Image trainer, checkpoint selection and independent inference reuse.

• Video reframe: source-preserving crop or Wan VACE outpainting

  Native FFmpeg crop/trim with no rescale and copied audio, or Wan 2.1 VACE expansion of moving footage (widen a vertical clip to 16:9, or the reverse) with a declared changed-pixel mask, restored source pixels and source audio. Clips over 81 frames are covered by overlapping passes, each conditioned on the previous pass's delivered frames. check-expansion measures any widened video, including one made elsewhere, for frozen or discontinuous bands.

• Brand logo watermark on a finished video

  Uploaded or Wavemaker-generated logo intake, workspace brand logo default, and one-pass FFmpeg logo overlay graph (WavemakerVideoWatermark) with destination-aware placement, opacity, timing and legibility shadow, producing a new video artifact.

• MelBand vocal isolation and voice-forward stem remixing

  Preserved five-node MelBand graph, exact installed checkpoint/runtime, two owned source trials, full-WAV review, targeted native remix and independent-project reuse.

• Reviewed native repairs and source decisions for 207 catalog originals

  108 explicit Comfy API derivatives with preserved originals, staged-input manifests and model dependencies; 99 orchestration shells consolidated into named native methods. See per-entry decisions and runtime evidence.

• Bounded native Comfy loops, state carry and recoverable batch crops

  Two ordinary pinned core graphs with ordered finite image variants and state carry, native baseline, actual failure/cancellation partial collection, explicit recovery, targeted iteration revision and independent package reuse.

• 401 native Comfy style prompts and portable recipes

  All 176 image and 225 video style entries as a standalone standard-library Comfy node pack. Exact brief, overridable guidance and provenance are ordinary STRING outputs.

• Portable Comfy VideoEdit crop and trim with audio

  Ordinary core VideoTrim/VideoCrop graphs with complete nested VideoEdit objects, preserved Wavemaker package round-trip, exact owned SDR8/CFR range/crop comparisons, parameter revision and independent execution reuse.

• Build arbitrary graphs from nodes, including one multi-scene video graph

  Author arbitrary Comfy API graphs from inspected nodes, model loaders and typed ports; validate and submit supported graphs. Shared character, clothing, setting and object plates can condition multiple scenes.

• Arbitrary native graphs, dependency preparation and bounded frontend import

  Byte-locked dependency preparation on an initially incomplete runtime, ordinary graph topology edits, package transfer, and bounded official frontend fixture conversion.

• Grounded product UGC promo with references and authored speech

  Editable native method for supplied product photos, recurring creator identity, reference-conditioned scene plates and H3 shots, exact narration/dialogue and timed finishing.

• Grounded live-action movie trailer: character continuity, narration and dialogue

  Editable multi-scene treatment: researched known characters, cast, objects and locations; reference-conditioned scene images; H3 reference shots; authored speech, voice and timed audio finishing. Ordinary graphs and reusable components, customized by the calling agent.

• Self-hosted image edit tools: erase, inpaint, expand, reframe, relight, restyle, background, sketch, face, depth

  Qwen-Image-Edit 2511 Lightning tool graphs and profiles ported from the legacy wm-recipes edit, enhance and make-images tools, plus FireRed-Image-Edit 1.1, Depth Anything 3 and LTX-2.3 IC-LoRA restore graphs. What each tool is good for, typical cost and time, and when to pick another.

• Background removal and matting: product, logo and people cutouts, hair, glass and video alpha

  BiRefNet (general, portrait, HR-matting, lite-2K, Lucida) and BRIA RMBG-2.0 graphs and profiles on the birefnet T4 fleet variants: RGBA cutout plus matte for stills, VP9 alpha WebM plus matte video for clips. Replaces the WavemakerCutout http (WM_CUTOUT_URL) backend. What each engine is good for, cost and time, and how to composite the result.

• Long film: one intact Comfy graph from 10 seconds to 60 minutes

  Author one Comfy graph: self-hosted keyframes, H3 shots written to file-backed VIDEO, scene reels, then a master via nested WavemakerConcatVideos. Short estimates use the interactive fleet lane; longer work uses the durable modal-film lane with comfy-cache-v1 checkpoints. Partner nodes are refused. The only product limit is the approved spend.

• LTX-2.3 IC-LoRA supplied-face video repair with a declared changed-pixel bound

  Reference-face identity keyframe, face-parsing bound, union-control IC-LoRA regeneration of masked face latents guided by source DWPose, and exact source pixels/audio outside the bound.

• Model strengths by job: which model measured best for stills, text, product, restyle, identity restage, music, voice and video

  A short dated reference of measured comparisons (bake-offs, blind reviews, ArcFace identity scores, OCR) naming the strongest and weakest models per job, with evidence paths. Knowledge for choosing a model, not a fixed recipe.

• Motion transfer on self-hosted H3: a character performs a driving clip's dance or action, or replaces the person in that video

  Kling-style motion control on H3 from one driving clip and one character still: Viggle-Animate (H3 ref2va finetune) character replacement from a repainted reference frame, with windowed chaining for long takes; H3 Fun ControlNet Union 2.0 driven by SDPose skeletons; and stock H3 Ref2VA with the clip as <Video 1> attribute_transfer. Includes clip conforming, reference-frame extraction, repaint, keypoint extraction, side-by-side sheets and a bone-direction pose-adherence score, plus the pinned runtime layer and weights. Generation profiles h3-motion-viggle, h3-motion-pose and h3-motion-prompt bind an owned driving clip and still through generation.prepare, which measures the clip (frames, grid, canvas, cuts) and refuses rather than conforms; self-hosted character (Krea2 Turbo) and repaint (Qwen-Image-Edit 2511) stills replace the Gemini steps, which stay as partner options.

• MiniMax Music3 INT8 instrumental beds and bounded lyric comparison

  Preserved native INT8 graph, actual shared H100 lyric and no-vocal EDM trials, full-audio review, failed targeted lyric revision, native peak/fade finishing and independent package reuse.

• Native Bria, Recraft styles, Muse, Tencent and SenseNova model ports

  Ordinary Comfy API authoring for exact Bria edits, Recraft raster/vector style creation and reuse, Muse/Tencent generation/editing and self-hosted SenseNova. Separate pinned current runtime and preserved source-era interfaces. Three SenseNova graph profiles completed bounded, visually reviewed private B200 trials.

• Durable long-film stages, recovery and localized revision

  A 397.708-second authored educational film, a 28.625-second authored French derivative, crash recovery and one-chapter revision preserving selected siblings.

• Reviewed identity reference acquisition and reuse

  Keep the same character across scenes using scoped reference discovery, reviewed immutable identities and consumed graph bindings.

• Headless 360-image rectilinear camera and repeatable framing

  Ordinary local-image Comfy graph with explicit numeric camera, owned landmark/seam/pole controls, native comparison, targeted FOV revision and fresh independent package reuse.