Adrian Arcega retweeted
I get almost zero bad generations now with Seedance 2.5... The key is to give the model exactly what it wants:
- Character sheet for everyone
- Multiple views of the setting so it can construct a space without guessing. I use a grid of 4 angles.
- Prompt formatted using the official Seedance 2.5 skill
Combine it all in you app of choice (I use @floraai )
- Important detail -
Upload the references IN THE SAME ORDER as the prompt (eg, if it says "Reference 1: Kyle", make sure kyle.jpg is the first thing you connect).
It takes some time to set all this up, but I get almost zero bad generations now. If I do re-roll, it's generally because I want to make a creative change, not because of an error.
Saves a lot of credits and frustration!
Adrian Arcega retweeted
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models
“I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that.
“With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor.
“You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something.
“That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’
“I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
$ANET Andy Bechtolsheim says AI has made optics demand about 10 times bigger in five years and the industry is only at the "very beginning"
"So I guess I don't need to tell you that AI has been driving this incredible increase in demand for high-speed optics, which is probably now 10 times bigger than it used to be five years ago..."
"And what I want to talk to you today is that we're not at the end of this journey, but rather the very beginning. There's easily another order of magnitude increase in bits needed for the next generation kind of data centers."
"So what I want to talk about first is what's happening at the data center level, and how we as an industry have to get together to solve these problems of this very rapid and high demand growth as fast as possible, since AI can't wait."
______
For full set of takeaways to Andy's AI Datacenter and Optics lecture: firesidealpha.substack.com/p…
MUSIC INDUSTRY PRO TIP - be nice to everyone!
You'll encounter a lot of different kind of people at different levels in your career.
Some will be junior.
Some will be more senior.
Some in the middle.
Don't just be nice to those in positions of power...
Be nice to the interns, entry level team members, etc.
Don't just be nice to the director...
Be nice to the PA, sound person, lighting person, etc.
Being a genuinely nice and good person is underrated and can take you far.
Adrian Arcega retweeted
Most AI filmmakers know the names of camera angles.
Very few know what those angles are actually doing to the story.
- A low angle is not automatically “cinematic.”
- A Dutch angle is not something you add because the frame feels boring.
- And a high angle does not simply mean “weak.”
The camera should have a reason to be where it is.
Here are 8 angles worth understanding:
→ Eye-level: For honesty, intimacy, dialogue and natural performance.
→ Low angle: For authority, threat, monumentality and pressure.
→ Ground-level: For speed, scale, physicality and extreme size contrast.
→ High angle: For vulnerability, isolation, exposure and entrapment.
→ Overhead / Top-down: For geography, choreography, patterns and surveillance.
→ Dutch angle: For instability, unease and moments when the story itself feels wrong.
→ POV / Subjective: For immersion, fear, urgency and restricted information.
→ Over-the-shoulder: For dialogue, eyelines, relationships and screen geography.
But this is the important part for AI filmmaking:
Do not prompt:
“Cinematic low-angle shot of a powerful man.”
Direct the geometry instead:
SHOT SIZE + CAMERA HEIGHT + ANGLE + LENS + COMPOSITION + CAMERA BEHAVIOR + ACTION
Example:
“Medium-wide low-angle shot, camera 50 cm above the floor, 32mm lens. Commander stands screen-center beneath converging ceiling beams. Prisoner remains seated in the foreground. Locked camera. The commander slowly lowers his eyes toward him.”
That gives the model something useful:
where the camera is,
what it sees,
how perspective should feel,
where the characters are,
and what happens during the shot.
The rule I keep coming back to:
Do not choose the angle that looks coolest.
Choose the angle that communicates the moment best.
I put all 8 angles, their storytelling purpose, scene examples and AI prompts into one cheat sheet.
Save it for your next AI film. Share it with someone who is still prompting “cinematic camera angle” and hoping for the best.
PS:) Follow for more!
.
.
Artists & managers: Spotify is giving eligible artists the ability to upload video content to your profile.
You can now upload:
-Music videos
-Live performances
-Visualizers
-& more to your profile
You can assign these visuals to specific tracks so within the app fans have the option of either listening to the audio or watching.
If you have music videos on YouTube, it can't hurt to upload to Spotify and see if they impact user retention at all.
I recommend this to all of the artists I work with as it's another way to repurpose content.
Created with @ChatGPT image 2.5
prompt :
Create a premium professional character design reference sheet / production model sheet based strictly on the provided reference image.
REFERENCE & IDENTITY LOCK:
Use the uploaded reference as the single source of truth for the character's identity. Preserve the exact facial identity, facial structure, hairstyle, hairline, eye shape and color, eyebrows, nose, lips, skin tone, body proportions, physique, age appearance, distinctive features, costume design, accessories, footwear, colors, patterns and all recognizable visual details. Do not redesign, beautify, simplify, age, de-age, or reinterpret the character.
CHARACTER DESIGN ANALYSIS:
Before constructing the sheet, internally analyze and lock the character's:
facial construction
head-to-body ratio
body proportions
shoulder width and torso structure
limb length and joint placement
silhouette
hairstyle and hair volume
costume construction
accessory placement
color relationships
material characteristics
distinctive identity anchors
All subsequent views must represent the exact same character design.
PAGE FORMAT:
Create one sophisticated studio-grade character reference board, landscape 16:9, clean neutral/off-white studio background, refined editorial production-board aesthetic, highly organized hierarchy, generous spacing, subtle technical guide lines, restrained professional typography-style labels where appropriate, no decorative clutter.
01 — HERO CHARACTER PORTRAIT
Place one larger polished head-and-shoulders or three-quarter portrait as the visual identity anchor.
Show the character's canonical facial identity clearly with neutral professional expression and accurate hairstyle, skin, costume and accessories.
02 — FULL-BODY TURNAROUND
Create a clearly organized full-body turnaround showing the same character at identical scale and proportions:
FRONT VIEW
3/4 FRONT VIEW
SIDE PROFILE
3/4 BACK VIEW
BACK VIEW
Use a neutral standing pose with consistent posture and anatomical alignment.
Keep head height, eye line, shoulder line, waist, hips, knees and feet consistently aligned across every view.
The costume, hairstyle, accessories, seams, patterns, footwear and silhouette must remain identical from every angle.
03 — EXPRESSION STUDY
Include a clean expression grid containing approximately 6 expressions:
Neutral
Happy / subtle smile
Serious
Angry / determined
Surprised
Sad / emotional
Maintain the exact same facial identity, head proportions, hairstyle and facial construction in every expression.
Expressions should demonstrate believable facial acting rather than exaggerated deformation.
04 — SIGNATURE POSE STUDIES
Include 4–6 full-body pose studies that communicate the character's personality and physical behavior.
Use varied but believable poses such as:
relaxed standing
confident stance
walking
sitting
interacting with an object
dynamic signature pose
Maintain exact character proportions, costume construction and recognizable silhouette in every pose.
05 — COSTUME & DETAIL CALLOUTS
Add several clean close-up detail panels showing the most important design elements:
hairstyle / hair detail
face detail
collar / neckline
sleeves / garment construction
footwear
jewelry or accessories
distinctive emblem / pattern / texture
important prop if present
Show construction and material clearly without turning the sheet into a decorative fashion collage.
06 — MATERIAL STUDIES
Visually communicate the primary materials present in the design:
fabric, leather, metal, denim, silk, knit, plastic, glass, jewelry, hair, skin or other relevant materials.
Show realistic surface behavior, texture, reflectivity and construction appropriate to each material.
07 — COLOR PALETTE
Include a compact professional color palette strip containing the dominant character colors.
Organize colors according to their visual role:
skin
hair
primary costume
secondary costume
accent color
accessories / materials
Keep the palette faithful to the reference.
08 — PROPORTION & SILHOUETTE GUIDE
Include a subtle technical proportion guide beside the turnaround.
Show:
overall height
head-to-body ratio
major horizontal alignment guides
key body proportions
clean silhouette thumbnail
Keep this section understated and production-oriented.
09 — DESIGN CONTINUITY
Treat the entire page as a single canonical character source of truth.
Every panel must depict the SAME person/character with:
identical facial identity
identical body proportions
identical hairstyle
identical costume
identical accessory placement
identical color palette
identical design language
consistent left/right details
No accidental costume changes, missing accessories, duplicated accessories, altered facial features, changing body proportions, inconsistent hairstyles or unexplained design variations.
VISUAL DIRECTION:
High-end professional character design presentation, studio production reference quality, sophisticated concept-art discipline, clean polished rendering, precise construction, controlled neutral lighting, realistic material definition, excellent anatomical consistency, crisp readable details, refined editorial layout, premium art-direction quality.
The sheet should feel like an actual professional animation / game / visual-development production document, not a collection of random AI images.
COMPOSITION:
Clear information hierarchy, balanced negative space, aligned panels, consistent character scale, clean grid system, logical visual flow, no overlapping figures, no cropped bodies, no confusing perspective, no unnecessary scenery.
CAMERA / VIEW CONTROL:
Turnaround views should use consistent orthographic-like framing and neutral perspective. Expression studies should use a consistent head framing. Pose studies may use natural perspective while preserving character proportions.
LIGHTING:
Neutral studio illumination designed for design inspection rather than cinematic drama. Soft, even, physically believable light with controlled shadows and accurate material readability.
PHOTOGRAPHIC / RENDER FINISH:
Ultra-clean high-end visual development presentation, realistic surface detail, natural skin and hair rendering, physically believable materials, sharp but refined detail, professional production-board finish.
OUTPUT QUALITY:
16K: 15360 × 8640 ≈ 132.7 million pixels, high-resolution professional quality.
NEGATIVE CONSTRAINTS:
No character redesign, no identity drift, no inconsistent proportions, no changing face, no changing hairstyle, no costume variations, no missing accessories, no duplicated accessories, no extra limbs, no malformed hands, no distorted anatomy, no random props, no dramatic scenery, no cinematic background, no excessive effects, no clutter, no watermark, no logo, no cropped views.
🤖 Made with AI
这应该就是以后影视剧制作的常态!!
就像我之前所说,现阶段对AI视频最好的用法就是:
实拍 + AI!
人物的情绪、动作、微表情,这些决定了内容的温度和质感,需要“绝对的真实”,依然采用实拍。
而光影、场景、VFX,氛围感塑造,都可以用AI来搞定。
用实拍去托住真实感,用AI去放大表达力!
视频来自新加坡首部AI混合剧情片:《The Crooks》
最近身边很多朋友都在讨论,AI会不会取代影视行业里的实拍?
但如果你真的做过内容、拍过片子、交付过客户,就会越来越清楚一件事:
视频行业真正高效、也最有竞争力的方式,不是纯AI,也不是死守纯实拍,而是AI+实拍的结合。
AI的优势,是快,是低成本,是能迅速把脑子里的想法“具象化”;它特别适合做概念预演、气氛镜头、复杂转场、难以实拍的想象性画面,也适合把一个普通团队的创意上限拔高。
但它的问题也很明显,人物真实感、情绪细节、品牌信任感,即便是最强的Sd2.0,其实还是差着那一口“活人气”。
而纯实拍的价值,恰恰就在这里!!
真实的人、真实的光线反馈、真实的微表情,这些东西决定了内容有没有温度,有没有说服力,有没有质感。
尤其是商业化内容,能够真正产生溢价的,往往不是炫技,而是信任,是人与人的情感互动。
所以未来最强的创作者,不是最会拍的人,也不是最会跑AI的人,而是最会把两者编排在一起的人。
用实拍去托住人物的真实感,用AI去放大表达力;把该真实的部分拍下来,把该夸张、该想象、该提升效率的部分交给AI。
说得更直白一点,AI不是来替代镜头的,AI是来解放镜头的。
它把那些过去要花很多钱、很多时间、很多人力才能完成的部分,比如特效、比如调色、比如光影氛围、比如布景、比如道具,变成了一个人也能调度和解决的事情。
但真正决定作品质感下限和商业价值上限的,依然是真实本身。
视频来自:边境悍匪
Here are 9 tools anyone in music could build for themselves without any prior coding experience:
1. Release metadata extractor - paste a link, pull every ISRC, UPC, and credit in 30 seconds (the one I built).
2. Royalty tracking dashboard - aggregate distributor statements, P&L's, etc all in one place.
3. Social media posting schedule planner - calendar of what is getting posted and when.
4. Playlist pitch tracker - who you pitched, who passed, who added, and when to follow up.
5. Tour routing tool - pull your streaming data by city and map the most efficient run of dates.
6. Split sheet generator - fill in collaborators and percentages, export a clean signed-ready PDF.
7. Sync brief matcher - paste a music supervisor's brief, get your best-fitting tracks ranked.
8. Contract red-flag scanner - drop in a deal memo and get plain-English flags before you sign. (note: this does not replace proper legal counsel).
9. CRM - track your contacts across Gmail and other communication sources.
Anyone can literally build almost anything.
What's crazy is I used to pay for a CRM tool to organize my contacts...fast forward and over the past 30 days, I just built my own using my own dashboard, catered to my specific needs.
"I wish a tool did this" became "I can build a tool that does this"!
Adrian Arcega retweeted
Handheld camera motion for Seedance 2.0
I hate blender so I stage it on my iphone.
You stage the scene (open source link below) and can walk around it and capture real camera movement in 3D space, then Seedance populates the scene.
Handheld camera motion and realistic shake is hard to describe or prompt to AI. The iphone’s ARKit is super precise with motion tracking and you can record complex scenes as a one man studio.
Here is the AI filmmaking workflow:
1. Set up blocking in the 3D space
2. Record camera motion
3. Generate a starting image with location and characters
4. Feed image and video to Seedance
This was a fun build with claude code. Happy if others want to build on it. – at Palo Alto, CA
Seedance 2.0 is insane.
Blender previs is having a moment in AI filmmaking for a reason.
I wanted to see how far I could push the workflow beyond a simple camera move in an ambush scene.
The workflow:
1. Make a start image in Midjourney
2. Block it out in Blender, animate the camera
3. Feed both to Seedance
The more I test this pipeline, the more convinced I am that hybrid pipelines are the future. The conversation is moving fast but previs is one of the clearest unlocks so far.
Seedance still isn't perfect at matching the reference but it's close.
You can experiment with staging, timing, and camera movement before you ever touch the video model.
Solve the scene before you generate the shot.
Adrian Arcega retweeted
Incredible to see what a professional animator can do with AI video.
He created a 3D previz and then had Seedance render the actual anime - with the motion and camera control preserved!
(made by @craftcapitallab)
Adrian Arcega retweeted
This is wild.
Every filmmaker needs to try this hybrid AI workflow for precision camera control:
1) Block out your camera in Blender to create a motion reference and animate the camera move
2) Generate your start frame in Midjourney to match the pose and composition
3) Feed both to Seedance
It tracks the move almost exactly.
This is the most powerful hybrid pipeline I've tried yet.
MUSIC INDUSTRY PRO TIP – The industry is smaller than you think.
Like...really really small lol.
Everyone knows each other and the intern today might be a power player tomorrow.
I've watched it happen multiple times.
How you treat people on the way up is how they’ll remember you when they’re up.
New path control system in Seedance 2.0 just cooked Hollywood...
- You draw the camera move by hand.
- Scribble a red line across a still image
- Feed it to the model and it flies the shot along that exact route.
Then it erases the line so it never shows up in the footage.
The marker stroke becomes the flight path.
Prompts below on @unsora_ai👇
MUSIC INDUSTRY HOT TAKE - how artists are tagged on DSPs is broken.
Artists would benefit from DSPs having a checks and balances system - when an artist is being tagged as a primary artist on a release that's not theirs, they must choose to accept it in order for the release to show up on their profile and in their DSP backend.
It's so easy for an artist to get tagged by another artist not realizing they are interrupting a rollout and potentially causing massive consequences, intefering with that artist's original release.
It's unfair to the artist and a broken system in general.
Adrian Arcega retweeted
This is pretty insane: the U.S. just tried to literally re-colonize part of the Philippines.
They did so under the so-called "Pax Silica" initiative, the brainchild of - surprise, surprise - an ex-Palantir guy named Jacob Helberg who now runs U.S. economic "diplomacy" from the State Department.
It's causing a big outcry in the Philippines, which is quite a feat given this is by far the most US-friendly country in Southeast Asia.
If you're the US and you're getting the Marcos administration - of all governments - to push back on sovereignty, you've really overplayed your hand.
What is the "Pax Silica" initiative? In a nutshell it's about the US getting other countries to commit to restructuring their AI tech infrastructure around a US-led stack. It's basically vendor lock-in: you hand over your critical minerals, align your export controls with Washington's, regulate AI the way America wants, and in return you get to be a US "trusted partner," whatever that means these days.
In essence, let's not kid ourselves, it's all about China: this is the US's initiative to "win the AI race" by getting other countries to contractually commit to keeping China out of their tech supply chains. When you can't preserve your lead through innovation, you seek to lock countries in contractually.
For instance as a country, this would mean telling Huawei they can't sell you AI chips, and telling Chinese firms they can't invest in your data centers - even if they're better and cheaper. It's not about choosing the best technology, it's about choosing the right flag.
But in this instance, the US went much further still: they literally tried to carve out 4,000 acres of Philippine territory (in New Clark City, 60 miles north of Manila) to be governed under US common law with diplomatic immunity - the first arrangement of its kind anywhere in the modern world.
This is according to the WSJ who ran the story last month (wsj.com/world/asia/u-s-to-cr…) as if it was a done deal (it wasn't).
Heard about the "French concession" or "British concession" in China during the century of humiliation? Same thing: the US basically asked for an "American concession" in the Philippines.
Unsurprisingly, there was quite a bit of backlash in the country with for instance the Peasant Movement of the Philippines (KMP) calling it a “massive sellout” of the country’s land, minerals, and sovereignty (punto.com.ph/us-led-pax-sili…).
So much so that the Philippines' government - namely Joshua Bingcang, president and chief executive of the Bases Conversion and Development Authority (BCDA) - issued a statement saying that the Philippines had rejected US proposals that would place the project beyond local jurisdiction (asianews.network/philippines…).
Note, by the way, this delicious irony: the BCDA is the government agency that was created in 1992 specifically to convert former US military bases at Clark and Subic Bay after the Philippines spent decades negotiating their closure. New Clark City - where the Pax Silica's hub would go - is built on the old Clark Air Base.
So the agency whose entire reason for existing is to turn former American colonial territory (i.e. US military bases) into sovereign Philippine land is the one now being asked to hand part of that very same land back under US jurisdiction (and, apparently, declined).
Of course though, blocking this specific jurisdiction grab doesn't change the bigger picture. The Philippines is still a Pax Silica signatory, and Pax Silica itself is structurally neocolonial: you supply the cheap labor and raw materials, align your export controls and regulations with Washington's, cut yourself off from the world's rising technological powerhouse - and in exchange you get assembly jobs and the privilege of getting a pat on the head and being called a "trusted partner."
They dropped the most cartoonishly colonial demand - governing Philippine soil under US law - but the underlying architecture is the same: you serve America's supply chain, on America's terms, and you relinquish your sovereign right to trade with whoever offers the best deal.
GPT Image 2 - Character Reference Sheet Prompt
You can create a character reference sheet in any style you want for use in your Seedance 2.0 videos. All you need to do is either provide a reference image or describe your character. You can also input random values to generate completely random characters.
This reference sheet provides a clear, structured, and highly readable visual breakdown of the character, helping the model maintain strong identity consistency across different outputs. The layout ensures the model focuses on essential visual information rather than stylistic noise, while the inclusion of expressions, head angles, and performance cues improves the model’s ability to generate believable acting and varied poses without losing character integrity.
All clips in the demonstration video were generated using the reference sheet image and the same Seedance 2.0 prompt. You can find the prompts in the replies. 👇
GPT Image 2 and Seedance 2.0 are both available on @mitte_ai
🤝 Paid partnership
Adrian Arcega retweeted
Wow... Someone built the Pinterest of AI image prompts and it's 100% free.
It's called MeiGen. A massive gallery of viral prompts for GPT Image 2, Nano Banana 2, Seedance 2.0, Veo 3.1, and Midjourney all in one place.
No prompt engineering. No paid courses. Just steal what already went viral.
100% free.



























