
Rafael Neves
ABOUT
A 1:42 AI cinematic showreel by Rafael Neves, senior video editor and AI filmmaker. One actor travels through ten worlds, from a zombie graveyard to a coffee-shop conversation where the twist is that everything, including the other actor, is AI. Built with Higgsfield Cinema Studio (Seedance 2.5) for video, Nano Banana Pro, Soul Cinema and GPT Image 2 for characters and sets, ElevenLabs for voice, and finished in Adobe Premiere Pro, After Effects and DaVinci Resolve. What it shows: • Character consistency across every block (locked character sheets and outfits) • Multi-shot scenes with planned camera language • Dialogue with lip-sync • A real edit, grade and sound mix, not raw generations Open to AI filmmaker, AI video editor and cinematic AI roles. Remote, UTC-3, US time-zone overlap. Portfolio: rafael-neves.myportfolio.com LinkedIn: linkedin.com/in/rafaellneves22 Email: rafaelnevesdesigner022@gmail.com
SELECTED FRAMES — 8 OF 8
PROCESS
BRIEF 1:42 self-pitch showreel for AI filmmaking work. One narrative showing the full stack: writing, character design, cinematography, VFX, sound, color, edit. 10 thematic blocks referencing directors (Snyder, Deakins, Villeneuve, Duffer, Fincher, Nolan). STACK Higgsfield Cinema Studio 4.0 (Seedance 2.5) for video. Nano Banana Pro, Soul Cinema, GPT Image 2, Genjutsu Object Swap for image and scene ops. Omni Flash for budget shots. ElevenLabs for voice. Premiere + Dehancer Pro (Kodak Vision3 500T), After Effects and DaVinci for grade and finish. ASSETS FIRST 4 characters, 6 outfits, 8 environments, 8 portals with one warm orange effect, chroma green props for post overlays. Every asset locked on mid-gray flat lighting before any video gen, so plates carry zero lighting info downstream. NUMBERS 50 video gens across 10 blocks. Frame-by-frame face swap on Block 10A when the Genjutsu queue stalled for 2+ days: Nano Banana Pro on frontals, GPT Image 2 on back views. LOOK Graded against Roger Deakins language: matte skin, atmospheric perspective, filmic roll-off, 35mm grain, halation on warm highlights. Kodak Vision3 500T applied uniformly. MUSIC 2 needle drops instead of underscore. The "Live and Let Die" orchestral drop lands on the first portal ignition. Hard cut into "Rocket Man" for the coffee-shop reveal. The tonal contrast is the joke. WHAT FAILED AND THE FIX • Seedance flips LEFT/RIGHT about half the time. Fix: state the direction three ways. • "Continuous single take" collapsed a multi-angle scene into one shot. Fix: explicit SHOT 1 / SHOT 2 blocks. • Content filters react to visual patterns, not words. Fix: original characters and generic silhouettes. • Slow motion never rendered at true speed. Fix: retime in the edit. • Nano Banana can't build back-view faces from a frontal ref. GPT Image 2 can.


