Emociones con IA 🥲
Hoy quise poner a prueba los mejores generadores de vídeo con IA para ver si de verdad son capaces de transmitir diferentes emociones 👀
Usé la misma imagen y el mismo prompt para generarlas, y aun así cada uno me da un resultado distinto… Os dejo los testeos que hice para que podáis juzgar vosotros mismos qué generador lo hace mejor 😋
Y, por cierto, mañana Kling lanza su nueva versión: Kling 3.0. Pronto tendréis nuevos vídeos poniéndolo a prueba
Y como siempre, si comentas “ARIA”, te paso todos los prompts de las imágenes y de las emociones que usé 💌
Why soy_aria_cruz's AI Emotion Test AI Video Went Viral and the Formula Behind It
An emotion benchmark is more useful than another pretty demo
This clip is a facial-performance benchmark wrapped as social content. The frame is brutally simple: one woman in a dark close-up, wearing large round glasses and hoop earrings, crying and moving into a full terror scream while the lower-third text labels the test as “GRITO DE TERROR” and “SEEDANCE 1.5 PRO.” There is no distracting background, no costume change, and almost no camera movement. That simplicity is exactly why the clip works. The whole post is asking a sharp creator question: can current AI video models actually sell emotion on a human face when everything depends on micro-expression, tears, jaw shape, and eye tension?
For creators, that is valuable because emotion shots are where weak models get exposed fast. A wide action clip can hide problems in motion blur, but a locked close-up cannot. This test turns a single face into a judge. The audience is not just watching for aesthetics. They are watching for believability.
What You're Seeing
The frame design
The composition is centered and intimate. The woman fills nearly the entire frame, which forces attention onto the emotional performance. The background stays dark and neutral, so the wetness under the eyes, the forehead tension, and the open mouth become the whole story. The glasses matter too. They are a difficult consistency object, and because they sit right on the face, any drift becomes obvious immediately.
What changes over time
The video does not rely on plot progression. It relies on intensity progression. The face moves from fear and crying into a more explicit terror scream, then sustains that state. That makes the clip a clean benchmark for emotion rendering rather than narrative storytelling.
Shot-by-shot breakdown
Time range
Visual content
Shot language
Lighting & color tone
Viewer intent
0:00-0:02 (estimated)
Tearful close-up, fear building in the eyes and forehead
Locked front-facing portrait
Dark olive-brown background, soft moody frontal light
Hook through realism and discomfort
0:02-0:04 (estimated)
Mouth opens wider, emotion escalates into panic
Same framing, tiny head movement only
Moisture highlights become more visible
Tests facial transition quality
0:04-0:06 (estimated)
Peak scream with stretched mouth and stronger facial tension
Extreme close-up performance hold
Low-key cinematic portrait lighting
Shows whether the model can sustain fear convincingly
0:06-0:08 (estimated)
Scream continues with small variations in mouth and eyes
No cut, no zoom, all focus on expression stability
Same dark benchmark look
Reveals consistency over time
Why this format is so revealing
Close-up emotion tests are hard because viewers know human faces too well. Even slight problems in lip shape, tear movement, or eye asymmetry feel wrong immediately. That is why this short clip is more informative than a broader cinematic montage. It strips away excuses.
Why It Went Viral
The topic connects to a real creator need
The caption says the creator used the same image and the same prompt across multiple video generators to compare emotion rendering. That is a strong setup because it answers a real question creators care about: not “which model is newest,” but “which model can actually perform.” Emotion is one of the hardest things to fake well, so the benchmark feels meaningful right away.
There is also a simple psychological reason this performs. Human faces pull attention, and extreme emotion pulls even more. The clip combines both. A crying, screaming close-up creates immediate discomfort and curiosity. Then the text overlay gives the viewer an interpretation frame: this is not random drama, it is a test. That reduces confusion and increases watch intent.
Platform-view analysis
From the platform side, this post is efficient because the first frame already contains the experiment. There is no intro card, no slow reveal, and no setup paragraph required. The audience sees the face, sees the tears, sees the benchmark label, and understands the premise. That makes retention easier and also encourages comments, because viewers can compare tools and argue about which result feels most human.
5 testable viral hypotheses
Observed evidence: the frame is a direct face close-up. Mechanism: faces stop scrolls faster than environmental shots. Replication: lead benchmark posts with the face, not the setup.
Observed evidence: the emotion is extreme and unambiguous. Mechanism: clear emotional states create stronger viewer reactions. Replication: test joy, terror, grief, and rage separately instead of vague “dramatic acting.”
Observed evidence: the lower-third text tells viewers which model they are seeing. Mechanism: labels reduce confusion and improve comparison behavior. Replication: name each tool directly in-frame when doing benchmark series.
Observed evidence: the caption frames this as the same prompt across multiple tools. Mechanism: fairness increases credibility. Replication: explain your test conditions clearly.
Observed evidence: the post teases an upcoming Kling 3.0 test. Mechanism: forward-looking creator discourse increases follows and return visits. Replication: use benchmark posts as chapters in an ongoing series.
How to Recreate It
How to run a clean emotion benchmark
Choose one face with strong readable features, especially eyes, brows, and mouth shape.
Lock the frame. A moving camera makes it harder to judge the model fairly.
Use a plain dark background so the expression carries the test.
Pick one named emotion at a time, such as terror, joy, sadness, or disgust.
Write the same emotional prompt for every model you compare.
Keep accessories like glasses or earrings if you want a tougher consistency challenge.
Export clips of similar length so viewers can compare fairly.
Label each clip clearly with the model name and emotion.
In the caption, tell viewers exactly what to judge: eyes, mouth, tears, or lip sync.
What to watch for when grading results
The best outputs will keep the eye focus stable, the glasses geometry clean, the tears believable, and the mouth interior natural during the scream. The weakest outputs usually fail in the teeth, lips, or asymmetry around the eyes.
Growth Playbook
3 opening hook lines
Same face, same prompt, different AI video model. Which one actually sells fear?
If a model cannot handle a close-up scream, it is not ready for serious acting shots.
This is the fastest way to test whether AI video can do real emotion yet.
4 caption templates
1. Hook: I tested the best AI video generators with the same image and the same emotion prompt. Value: this one is Seedance 1.5 Pro doing a terror scream. Question: Does it feel real to you? CTA: Comment ARIA for the prompts.
2. Hook: Pretty visuals are easy, believable emotion is harder. Value: close-up face tests expose the truth fast. Question: Which model should I compare next? CTA: I'll post the series.
3. Hook: This is how I benchmark AI acting performance. Value: I keep the frame simple so the face has nowhere to hide. Question: Would you rank this above Kling or below it? CTA: Tell me why.
4. Hook: Tomorrow Kling 3.0 drops, so I wanted a clean baseline first. Value: this terror test shows what current tools can and cannot do. Question: What emotion should I test next? CTA: Drop one word below.