TGW Studios

One million+ views in two weeks. One person. Four hours. A 52-second video from our cinematic cyberpunk series The Great Revolt landed across three platforms in two weeks: TikTok: 1,000,000+ views | 2,335 new followers Instagram: 287,000+ views | 2,273 new followers Facebook: 68,000+ views | 1,600 new followers --------------------------- Total: 1,355,000+ views | 6,208 new followers One asset, three platforms. Tools used: Midjourney for stills Anthropic Claude for scripting Runway Seedance 2 for video ElevenLabs for narration Adobe Premiere for the edit This is Part I of "The Great Wall," a piece in the Lore format of our series. The Lore videos are world-building pieces from inside Ad Kôrun, the megacity at the center of The Great Revolt. Story does the work. Always has, always will. The medium keeps changing: oral tradition, written word, theater, cinema, animation, AI-generated film. What holds an audience does not. A world built with intention. Characters worth following. A reason to keep watching, no matter how long the form, no matter how the work was made. The tools change the production. The story is why anyone stays. #IndependentFilmmaking #Cyberpunk #Worldbuilding #IndieIP #SoloCreator

Finally got my hands on Seedance 2 through Martini. First test. The pipeline: took an old Midjourney v7 concept image, ran it through FLUX 2 Max on the Black Forest Labs Playground for a photorealism pass, then fed that single frame into #Seedance2 with a short action prompt. What came back was three distinct camera cuts - her spinning around, grabbing the handlebar, riding off into the rain. From one still image. The way it interprets cinematography and cuts from a text prompt is ahead of anything I've tested so far. Not perfect - but the shot composition and pacing surprised me. One input image. One short prompt. Multiple coherent shots with actual camera logic. First test, and it already does this. Curious where this goes. - Jaws #Seedance2 ByteDance Martini Black Forest Labs Midjourney #FLUX2 #GenAI #AIvideo #ImageToVideo

#FLUX2 + Maxon #Cinema4D pipeline test. I quickly whipped up a cyberpunk cityscape as a gray-box blockout in Cinema 4D this morning. Rough geometry, basic composition, placeholder figures (sorry about the lazy mannequins 😂). Then took a viewport screenshot and ran it through FLUX 2 Max on the Black Forest Labs playground three different ways: 1. C4D viewport + a Midjourney style reference - gritty vertical megacity, rusted steel, neon through smog. The model picked up the composition from the viewport and the visual language from the MJ ref. 2. C4D viewport + prompt only - dark rain-soaked alley, cracked asphalt, neon kanji through haze. No style reference, just the prompt doing the work. 3. C4D viewport + prompt only - completely different direction this time. Pastel utopia. Cherry blossoms, marble streets, golden hour fog. Same geometry underneath, totally different world on top. Three completely different looks from the same scene in minutes. The point is the iteration speed - you can explore wildly different directions before committing to anything. Next step: animating the gritty cyberpunk version with LTX 2.3 by Lightricks in ComfyUI It's kind of wild how well 3D and AI workflows combine with this method. Imagine the use cases for this - creating a scene like this purely in C4D including shading and lighting would have taken days, not to mention rendertime. - Jaws Black Forest Labs Midjourney Lightricks #Cinema4D #FLUX2 #FLUX2Max #GenAI #3Dart #LookDev #AIart

Another quick test on Lightricks LTX 2.3. Started with an old Midjourney image, ran it through FLUX (Max) to push the look toward photorealism, then brought it straight into ComfyUI for LTX 2.3 image-to-video. The prompt adherence on this thing is really good. I ran the same scene through Veo 3.1 and honestly, LTX came out better on this one - both in quality and how closely it followed the prompt. That's worth sitting with for a second. A local model outperforming a cloud-based one. You do see a slight drop in photorealism as the camera moves, but that's expected from these video models right now, especially in a sci-fi setting (and not using first and last frame). Still, the overall quality is there. Planning to show proper side-by-sides eventually, but for now - just a quick look. - Jaws #LTX #LTX23 #ComfyUI #GenAI #AIvideo #Lightricks #Veo3 #Midjourney

Quick test on Lightricks' new #LTX 2.3, running locally on ComfyUI with a 4090. The quality caught me off guard. For a local model this is surprisingly close to Veo 3.1 territory - voices, music, overall coherence. There are imperfections, sure, but the fact that this runs on your own hardware without any API or cloud dependency is worth paying attention to. - Jaws #LTX #LTX2 #ComfyUI #GenAI #AIvideo #Lightricks

I thought Google #Gemini doesn't allow copyrighted material. Yet here we are. I'm running a stress test across the full Black Forest Labs FLUX 2 lineup (Max, Pro, Flex, Klein 9B, Klein 4B) and Google's #NanoBananaPro and #NanoBanana2. Same prompts, every model, two prompt formats - natural language and structured JSON. First up: The Power Broker. FLUX 2 Max is the real deal on this one. The JSON output came back with dust particles in the air, desk reflections, skin texture you can almost feel - and that SSS. The gap between Max and Pro is smaller than I expected, and Flex sits comfortably in the middle (although less photorealistic). Klein 4B is where you start feeling the compression, but even that is usable depending on context. Now the interesting part. The prompt describes a man in his late 50s behind a desk. No name. No reference to anyone specific. Just a fictional character description. Every single #FLUX output gave me an original, unique face. The #NanoBanana models? Swipe to slide 2 and decide for yourself. To be fair - Nano Banana did not exclusively produce celebrity lookalikes. Most outputs were original faces. But roughly a quarter of the runs came back with people you'd immediately recognize from a screen, and I picked the most obvious ones here because they were too striking to ignore. This is something I've noticed with Nano Banana before, not just on this prompt. (If anyone wants to see the non-celebrity outputs for a direct comparison, happy to share.) Google won't let you upload a celebrity's face to their image tools. But apparently the model has no problem generating recognizable faces on its own from a generic description. Interesting timing, given the whole #Seedance2 conversation with ByteDance right now. What do you see in the Nano Banana outputs? Curious what others think. - Jaws #GenAI #AIart #FLUX2 #NanoBanana #BlackForestLabs #AIImageGeneration #Flux2Max

One character. Two tools. Three camera angles. I took a Midjourney portrait reference and ran the same prompt through Black Forest Labs Flux 2 Max and #NanoBananaPro: "Move the camera to [angle], revealing new aspects of the scene while maintaining subject and environment consistency." Side portrait, wide establishing, extreme close-up. Simple instructions, no extra context. Basically Flux's own template. A few things stood out: The wide establishing shot is where it gets interesting. Flux played it safe and stayed relatively close to the original framing. NanoBanana pulled the camera all the way back and hallucinated an entire art studio with easels, paint supplies, sketches on the wall. Neither was "wrong" (I gave no environment direction), but the creative interpretation gap is massive. On the macro close-ups, both tools are producing skin detail that would have been impossible a year ago. Pores, individual hair strands, the wet sheen on the iris. At this distance, the quality difference is marginal. Both are scary good. On consistency: the character stays recognizable across angles in both tools, but neither nails it perfectly. Hair changes slightly, face shape drifts, features shift between generations. This is still the core problem with image-to-image workflows, and it's exactly what matters most for production work. Worth noting: I didn't optimize these prompts at all. This was a raw, out-of-the-box test. There's room to push both tools further with better direction. Swipe through for the full comparison. -Jaws #Flux #NanoBanana #GenAI #AIart #CharacterDesign